Machine learning algorithms for outcome prediction in (chemo)radiotherapy: An empirical comparison of classifiers

The value of PROMs for predicting erectile dysfunction in prostate cancer patients with Bayesian network

Purpose

This study aims to develop and externally validate a clinically plausible Bayesian network structure to predict one-year erectile dysfunction in prostate cancer patients by combining expert knowledge with evidence from data using clinical and Patient-reported outcome measures (PROMs) data. In addition, compare and contrast structures that stem from PROM information and routine clinical data.

Summary of background

For men with localized prostate cancer, choosing the optimal treatment can be challenging since each option comes with different side effects, such as erectile dysfunction, which negatively impacts their quality of life. Our previous findings demonstrate that logistic regression models are able to identify patients at high risk of erectile dysfunction. However, methods such as Bayesian networks may be more successful, as they intricately represent the causal relations between the variables.

Patients and methods

946 prostate cancer patients from 65 Dutch hospitals were considered to develop the Bayesian network structure. Continuous variables were discretized before analysis based on expert opinions and literature. Patients with missing information and variables with more than 25% of missing information were excluded. Prostate cancer treating physicians first determined the relationships (arcs) between the available variables. The structures were then modified based on algorithmically derived structures using the hill-climbing algorithm. Structural Performance was evaluated based on the area under the curve (AUC) values and calibration plots on the training and test data.

Results

BMI and prostate volume via MRI were excluded from this analysis due to their high percentage of missingness (>45 %). The final cohort was reduced to 505 and 216 after excluding 157 and 68 patients with missing information, respectively. The AUC of the PROM structure was better than the clinical structure in both the train and test data. The structure that combined both sources of information had an AUC value of 0.94 (0.92 - 0.96) and 0.84171 (0.77 91) in the train and test data, respectively.

Conclusion

Bayesian network structures derived from PROM information by complimenting expert knowledge with evidence from the data produce a clinically plausible structure that is more performant than structures from clinical data. Our study supports the growing global recognition of incorporating the patient's perspective in outcomes research for better decision-making and optimal outcomes. However, a structure that combines both sources of information gives a more holistic view of the patient with actionable insights and improved discriminative power.

Autophagy in Multiple Sclerosis: Phagocytosis and Autophagy of Oligodendrocyte Precursor Cells

Multiple sclerosis (MS) is a leading cause of chronic neurological dysfunction in young to middle-aged adults, affecting approximately 2.5 million people worldwide. It is characterized by inflammation, multifocal demyelination, axonal loss, and white and gray matter gliosis. Autophagy is a highly conserved protein degradation pathway. Polymorphisms in autophagy-related genes have been implicated in a variety of autoimmune diseases, including systemic lupus erythematosus, rheumatoid arthritis, psoriasis and MS. However, the significance of autophagy in MS remains to be elucidated. This paper aims to explore the potential role of autophagy-related genes in MS diseases by using bioinformatics combined with machine learning methods. Finally, we obtained 9 autophagy genes with the highest correlation with MS, and further changes in these autophagy genes were verified in the experimental autoimmune encephalomyelitis (EAE) model and oligodendrocyte precursor cells (OPCs) engulfed myelin debris (MD). Combined with bioinformatic analysis and experimental data, Becn1 showed obvious expression abnormalities suggesting that this gene has vital functions in autophagy and MD engulfed by OPCs. This work will be of great significance for the further exploration of autophagy-related genes in demyelinating diseases.

Evaluating the significance of ECSCR in the diagnosis of ulcerative colitis and drug efficacy assessment

Background

The main challenge in diagnosing and treating ulcerative colitis (UC) has prompted this study to discover useful biomarkers and understand the underlying molecular mechanisms.

Methods

In this study, transcriptomic data from intestinal mucosal biopsies underwent Robust Rank Aggregation (RRA) analysis to identify differential genes. These genes intersected with UC key genes from Weighted Gene Co-expression Network Analysis (WGCNA). Machine learning identified UC signature genes, aiding predictive model development. Validation involved external data for diagnostic, progression, and drug efficacy assessment, along with ELISA testing of clinical serum samples.

Results

RRA integrative analysis identified 251 up-regulated and 211 down-regulated DEGs intersecting with key UC genes in WGCNA, yielding 212 key DEGs. Subsequently, five UC signature biomarkers were identified by machine learning based on the key DEGs-THY1, SLC6A14, ECSCR, FAP, and GPR109B. A logistic regression model incorporating these five genes was constructed. The AUC values for the model set and internal validation data were 0.995 and 0.959, respectively. Mechanistically, activation of the IL-17 signaling pathway, TNF signaling pathway, PI3K-Akt signaling pathway in UC was indicated by KEGG and GSVA analyses, which were positively correlated with the signature biomarkers. Additionally, the expression of the signature biomarkers was strongly correlated with various UC types and drug efficacy in different datasets. Notably, ECSCR was found to be upregulated in UC serum and exhibited a positive correlation with neutrophil levels in UC patients.

Conclusions

THY1, SLC6A14, ECSCR, FAP, and GPR109B can serve as potential biomarkers of UC and are closely related to signaling pathways associated with UC progression. The discovery of these markers provides valuable information for understanding the molecular mechanisms of UC.

STAT3-Mediated Ferroptosis is Involved in Sepsis-Associated Acute Respiratory Distress Syndrome

Sepsis-induced acute respiratory distress syndrome (ARDS) poses a grave danger to life, resulting from sepsis-induced multi-organ failure. Although ferroptosis, a form of iron-dependent lipid peroxidative cell death, has been associated with sepsis-induced ARDS, the specific mechanisms are not fully understood. In this study, we utilized WGCNA, PPI, friends analysis, and six machine learning techniques (Lasso, SVM, RFB, XGBoost, AdaBoost, and LightGBM) to pinpoint STAT3 as a potential diagnostic marker. A significant increase in monocyte and neutrophil levels was observed in patients with sepsis-induced ARDS, as revealed by immune infiltration analyses, when compared to controls. Moreover, there was a positive correlation between STAT3 expression and the level of infiltration. Single-cell analysis uncovered a notable disparity in B-cell expression between sepsis and sepsis-induced ARDS. Furthermore, in vitro experiments using LPS-treated human bronchial epithelial cells (BEAS-2B) and THP1 cells demonstrated a significant increase in STAT3 phosphorylation expression. Additionally, the inhibition of STAT3 phosphorylation by Stattic effectively prevented LPS-induced ferroptosis in both BEAS-2B and THP1 cells. This indicates that the activation of STAT3 phosphorylation promotes ferroptosis in human bronchial epithelial cells in response to LPS. In summary, this research has discovered and confirmed STAT3 as a potential biomarker for the diagnosis and treatment of sepsis-induced ARDS.

Systematic review and meta-analysis of deep learning applications in computed tomography lung cancer segmentation

BACKGROUND

Accurate segmentation of lung tumors on chest computed tomography (CT) scans is crucial for effective diagnosis and treatment planning. Deep Learning (DL) has emerged as a promising tool in medical imaging, particularly for lung cancer segmentation. However, its efficacy across different clinical settings and tumor stages remains variable.

METHODS

We conducted a comprehensive search of PubMed, Embase, and Web of Science until November 7, 2023. We assessed the quality of these studies by using the Checklist for Artificial Intelligence in Medical Imaging and the Quality Assessment of Diagnostic Accuracy Studies-2 tools. This analysis included data from various clinical settings and stages of lung cancer. Key performance metrics, such as the Dice similarity coefficient, were pooled, and factors affecting algorithm performance, such as clinical setting, algorithm type, and image processing techniques, were examined.

RESULTS

Our analysis of 37 studies revealed a pooled Dice score of 79 % (95 % CI: 76 %-83 %), indicating moderate accuracy. Radiotherapy studies had a slightly lower score of 78 % (95 % CI: 74 %-82 %). A temporal increase was noted, with recent studies (post-2022) showing improvement from 75 % (95 % CI: 70 %-81 %). to 82 % (95 % CI: 81 %-84 %). Key factors affecting performance included algorithm type, resolution adjustment, and image cropping. QUADAS-2 assessments identified ambiguous risks in 78 % of studies due to data interval omissions and concerns about generalizability in 8 % due to nodule size exclusions, and CLAIM criteria highlighted areas for improvement, with an average score of 27.24 out of 42.

CONCLUSION

This meta-analysis demonstrates DL algorithms' promising but varied efficacy in lung cancer segmentation, particularly higher efficacy noted in early stages. The results highlight the critical need for continued development of tailored DL models to improve segmentation accuracy across diverse clinical settings, especially in advanced cancer stages with greater challenges. As recent studies demonstrate, ongoing advancements in algorithmic approaches are crucial for future applications.

Baseline DSB repair prediction of chronic rare Grade ≥ 3 toxicities induced by radiotherapy using classification algorithms

Small fractions of patients suffer from radiotherapy late severe adverse events (AEs Grade ≥ 3), which are usually irreversible and badly affect their quality of life. A novel functional DNA repair assay characterizing several steps of double-strand break (DSB) repair mechanisms was used. DNA repair activities of peripheral blood mononuclear cells were monitored for 1 week using NEXT-SPOT assay in 177 breast and prostate cancer patients. Only seven patients had Grade ≥ 3 AEs, 6 months after radiotherapy initiation. The machine learning method established the importance of variables among demographic, clinical and DNA repair data. The most relevant ones, all related to DNA repair, were employed to build a predictor. Predictors constructed with random forest and minimum bounding sphere predicted late Grade ≥ 3 AEs with a sensitivity of 100% and specificity of 77.17 and 86.22%, respectively. This multiplex functional approach strongly supports a dominant role for DSB repair in the development of chronic AEs. It also showed that affected patients share specific features related to functional aspects of DSB repair. This strategy may be suitable for routine clinical analysis and paves the way for modelling DSB repair associated with severe AEs induced by radiotherapy.

Pyroptosis-related long-noncoding RNA signature predicting survival and immunotherapy efficacy in patients with lung squamous cell carcinoma

Pyroptosis-related long-noncoding RNAs (PRlncRNAs) play an important role in cancer progression. However, their role in lung squamous cell carcinoma (LUSC) is unclear. A risk model was constructed using the least absolute shrinkage and selection operator (LASSO) Cox regression analysis based on RNA sequencing data from The Cancer Genome Atlas database. The LUSC cohort was divided into high- and low-risk groups based on the median risk score. For the prognostic value of the model, the Kaplan-Meier analysis, log-rank test, and Cox regression analysis were performed. A nomogram was constructed to predict the prognosis of patients, using a risk score and clinical parameters such as age, sex, clinical stage, and tumor node metastasis classification (TNM) stage. Afterwards, six common algorithms were employed to assess the invasion of immune cells. The Gene Set Enrichment Analysis (GSEA) was conducted to identify differences between patients at high and low risk. Furthermore, the pRRophetic package was employed to forecast the half-maximal inhibitory doses of prevalent chemotherapeutic drugs, while the tumor immune dysfunction and exclusion score was computed to anticipate the response to immunotherapy. The expression levels of the seven PRlncRNAs were examined in both LUSC and normal lung epithelial cell lines using RT-qPCR. Proliferation, migration, and invasion assays were also carried out to investigate the role of MIR193BHG in LUSC cells. Patients in the low-risk group showed prolonged survival in the total cohort or subgroup analysis. The Cox regression analysis showed that the risk model could act as an independent prognostic factor for patients with LUSC. The results of GSEA analysis revealed that the high-risk group showed enrichment of cytokine pathways, Janus tyrosine kinase/signal transducer and activator of the transcription signalling pathway, and Toll-like receptor pathway. Conversely, the low-risk group showed enrichment of several gene repair pathways. Furthermore, the risk score was positively correlated with immune cell infiltration. Moreover, patients in the high-risk category showed reduced responsiveness to conventional chemotherapeutic medications and immunotherapy. The majority of the long noncoding RNAs in the risk model were confirmed to be overexpressed in LUSC cell lines compared to normal lung epithelial cell lines by in vitro tests. Further studies have shown that downregulating the expression of MIR193BHG may inhibit the growth, movement, and infiltration capabilities of LUSC cells, whereas increasing the expression of MIR193BHG could enhance these malignant tendencies. This study found that PRlncRNAs were linked to the prognosis of LUSC patients. The risk model, evaluated across various clinical parameters and treatment modalities, shows potential as a future reference for clinical applications.

Advances and residual knowledge gaps in the neck management of head and neck squamous cell carcinoma patients with advanced nodal disease undergoing definitive (chemo)radiotherapy for their primary

PURPOSE

Substantial changes have been made in the neck management of patients with head and neck squamous cell carcinomas (HNSCC) in the past century. These have been fostered by changes in cancer epidemiology and technological progress in imaging, surgery, or radiotherapy, as well as disruptive concepts in oncology. We aimed to review changes in nodal management, with a focus on HNSCC patients with nodal involvement (cN+) undergoing (chemo)radiotherapy.

METHODS

A narrative review was conducted to review current advances and address knowledge gaps in the multidisciplinary management of the cN+ neck in the context of (chemo)radiotherapy.

RESULTS

Metastatic neck nodes are associated with poorer prognosis and poorer response to radiotherapy, and have therefore been systematically treated by surgery. Radical neck dissection (ND) has gradually evolved toward more personalized and less morbid approaches, i.e., from functional to selective ND. Omission of ND has been made feasible by use of positron-emission tomography/computed tomography to monitor the radiation response in cN+ patients. Human papillomavirus-driven oropharyngeal cancers and their cystic nodes have shown dramatically better prognosis than tobacco-related cancers, justifying a specific prognostic classification (AJCC) creation. Finally, considering the role of lymph nodes in anti-tumor immunity, de-escalation of ND and prophylactic nodal irradiation in combination are intense areas of investigation. However, the management of bulky cN3 disease remains an issue, as aggressive multidisciplinary strategies or innovative combined treatments have not yet significantly improved their prognosis.

CONCLUSION

Personalized neck management is an increasingly important aspect of the overall therapeutic strategies in cN+ HNSCC.

Predicting the Severity and Discharge Prognosis of Traumatic Brain Injury Based on Intracranial Pressure Data Using Machine Learning Algorithms

OBJECTIVE

This study aimed to explore the potential of employing machine learning algorithms based on intracranial pressure (ICP), ICP-derived parameters, and their complexity to predict the severity and short-term prognosis of traumatic brain injury (TBI).

METHODS

A single-center prospectively collected cohort of neurosurgical intensive care unit admissions was analyzed. We extracted ICP-related data within the first 6 hours and processed them using complex algorithms. To indicate TBI severity and short-term prognosis, Glasgow Coma Scale score on the first postoperative day and Glasgow Outcome Scale-Extended score at discharge were used as binary outcome variables. A univariate logistic regression model was developed to predict TBI severity using only mean ICP values. Subsequently, 3 multivariate Random Forest (RF) models were constructed using different combinations of mean and complexity metrics of ICP-related data. To avoid overfitting, five-fold cross-validations were performed. Finally, the best-performing multivariate RF model was used to predict patients' discharge Glasgow Outcome Scale-Extended score.

RESULTS

The logistic regression model exhibited limited predictive ability with an area under the curve (AUC) of 0.558. Among multivariate models, the RF model, combining the mean and complexity metrics of ICP-related data, achieved the most robust ability with an AUC of 0.815. Finally, in terms of predicting discharge Glasgow Outcome Scale-Extended score, this model had a consistent performance with an AUC of 0.822. Cross-validation analysis confirmed the performance.

CONCLUSIONS

This study demonstrates the clinical utility of the RF model, which integrates the mean and complexity metrics of ICP data, in accurately predicting the TBI severity and short-term prognosis.

Identifying lncRNAs and mRNAs related to survival of NSCLC based on bioinformatic analysis and machine learning

Non-small cell lung cancer (NSCLC) is the most common histopathological type, and it is purposeful for screening potential prognostic biomarkers for NSCLC. This study aims to identify the lncRNAs and mRNAs related to survival of non-small cell lung cancer (NSCLC). The expression profile data of lung adenocarcinoma and lung squamous cell carcinoma were downloaded in The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) dataset. A total of eight survival related long non-coding RNAs (lncRNAs) and 262 survival related mRNAs were filtered. By gene set enrichment analysis, 17 significantly correlated Gene Ontology signal pathways and 14 Kyoto Encyclopedia of Genes and Genomes signal pathways were screened. Based on the clinical survival and prognosis information of the samples, we screened eight lncRNAs and 193 mRNAs by single factor Cox regression analysis. Further single and multifactor Cox regression analysis were performed, 30 independent prognostication-related mRNAs were obtained. The PPI network was further constructed. We then performed the machine learning algorithms (Least absolute shrinkage and selection operator, Recursive feature elimination, and Random forest) to screen the optimized DEGs combination, and a total of 17 overlapping mRNAs were obtained. Based on the 17 characteristic mRNAs obtained, we firstly built a Nomogram prediction model, and the ROC values of training set and testing set were 0.835 and 0.767, respectively. By overlapping the 17 characteristic mRNAs and PPI network hub genes, three genes were obtained: CDC6, CEP55, TYMS, which were considered as key factors associated with survival of NSCLC. The in vitro experiments were performed to examine the effect of CDC6, CEP55, and TYMS on NSCLC cells. Finally, the lncRNAs-mRNAs networks were constructed.

A deep learning-based method for the prediction of temporal lobe injury in patients with nasopharyngeal carcinoma

PURPOSE

To establish a deep learning-based model to predict radiotherapy-induced temporal lobe injury (TLI).

MATERIALS AND METHODS

Spatial features of dose distribution within the temporal lobe were extracted using both the three-dimensional convolution (C3D) network and the dosiomics method. The Minimal Redundancy-Maximal-Relevance (mRMR) method was employed to rank the extracted features and select the most relevant ones. Four machine learning (ML) classifiers, including logistic regression (LR), k-nearest neighbors (kNN), support vector machines (SVM) and random forest (RF), were used to establish prediction models. Nested sampling and hyperparameter tuning methods were applied to train and validate the prediction models. For comparison, a prediction model base on the conventional D0.5cc of the temporal lobe obtained from dose volume (DV) histogram was established. The area under the receiver operating characteristic (ROC) curve (AUC) was utilized to compare the predictive performance of the different models.

RESULTS

A total of 127 nasopharyngeal carcinoma (NPC) patients were included in the study. In the model based on C3D deep learning features, the highest AUC value of 0.843 was achieved with 5 features. For the dosiomics features model, the highest AUC value of 0.715 was attained with 1 feature. Both of these models demonstrated superior performance compared to the prediction model based on DV parameters, which yielded an AUC of 0.695.

CONCLUSION

The prediction model utilizing C3D deep learning features outperformed models based on dosiomics features or traditional parameters in predicting the onset of TLI. This approach holds promise for predicting radiation-induced toxicities and guide individualized radiotherapy.

Machine Learning in Diagnosis and Prognosis of Lung Cancer by PET-CT

As a disease with high morbidity and high mortality, lung cancer has seriously harmed people's health. Therefore, early diagnosis and treatment are more important. PET/CT is usually used to obtain the early diagnosis, staging, and curative effect evaluation of tumors, especially lung cancer, due to the heterogeneity of tumors and the differences in artificial image interpretation and other reasons, it also fails to entirely reflect the real situation of tumors. Artificial intelligence (AI) has been applied to all aspects of life. Machine learning (ML) is one of the important ways to realize AI. With the help of the ML method used by PET/CT imaging technology, there are many studies in the diagnosis and treatment of lung cancer. This article summarizes the application progress of ML based on PET/CT in lung cancer, in order to better serve the clinical. In this study, we searched PubMed using machine learning, lung cancer, and PET/CT as keywords to find relevant articles in the past 5 years or more. We found that PET/CT-based ML approaches have achieved significant results in the detection, delineation, classification of pathology, molecular subtyping, staging, and response assessment with survival and prognosis of lung cancer, which can provide clinicians a powerful tool to support and assist in critical daily clinical decisions. However, ML has some shortcomings such as slightly poor repeatability and reliability.

Serum micro-RNAs with mutation-targeted RNA modification: a potent cancer detection tool constructed using an optimized machine learning workflow

RNA modifications affect fundamental biological processes and diseases and are a research hotspot. Several micro-RNAs (miRNAs) exhibit genetic variant-targeted RNA modifications that can greatly alter their biofunctions and influence their effect on cancer. Therefore, the potential role of these miRNAs in cancer can be implicated in new prevention and treatment strategies. In this study, we determined whether RMvar-related miRNAs were closely associated with tumorigenesis and identified cancer-specific signatures based on these miRNAs with variants targeting RNA modifications using an optimized machine learning workflow. An effective machine learning workflow, combining least absolute shrinkage and selection operator analyses, recursive feature elimination, and nine types of machine learning algorithms, was used to screen candidate miRNAs from 504 serum RMvar-related miRNAs and construct a diagnostic signature for cancer detection based on 43,047 clinical samples (with an area under the curve value of 0.998, specificity of 93.1%, and sensitivity of 99.3% in the validation cohort). This signature demonstrated a satisfactory diagnostic performance for certain cancers and different conditions, including distinguishing early-stage tumors. Our study revealed the close relationship between RMvar-related miRNAs and tumors and proposed an effective cancer screening tool.

Identification of GMFG as a novel biomarker in IgA nephropathy based on comprehensive bioinformatics analysis

Background

IgA nephropathy (IgAN) stands as the most prevalent form of glomerulonephritis and ranks among the leading causes of end-stage renal disease worldwide. Regrettably, we continue to grapple with the absence of dependable diagnostic markers and specific therapeutic agents for IgAN. Therefore, this study endeavors to explore novel biomarkers and potential therapeutic targets in IgAN, while also considering their relevance in the context of tumors.

Methods

We gathered IgAN datasets from the Gene Expression Omnibus (GEO) database. Subsequently, leveraging these datasets, we conducted an array of analyses, encompassing differential gene expression, weighted gene co-expression network analysis (WGCNA), machine learning, receiver operator characteristic (ROC) curve analysis, gene expression validation, clinical correlations, and immune infiltration. Finally, we carried out pan-cancer analysis based on hub gene.

Results

We obtained 1391 differentially expressed genes (DEGs) in GSE93798 and 783 DGEs in GSE14795, respectively. identifying 69 common genes for further investigation. Subsequently, GMFG was identified the hub gene based on machine learning. In the verification set and the training set, the GMFG was higher in the IgAN group than in the healthy group and all of the GMFG area under the curve (AUC) was more 0.8. In addition, GMFG has a close relationship with the prognosis of malignancies and a range of immune cells.

Conclusions

Our study suggests that GMFG could serve as a promising novel biomarker and potential therapeutic target for both IgAN and certain types of tumors.

A machine learning radiomics model based on bpMRI to predict bone metastasis in newly diagnosed prostate cancer patients

OBJECTIVES

To develop and evaluate a machine learning radiomics model based on biparametric magnetic resonance imaging MRI (bpMRI) to predict bone metastasis (BM) status in newly diagnosed prostate cancer (PCa) patients.

METHODS

We retrospectively analyzed bpMRI scans of PCa patients from multiple centers between January 2016 and October 2021. 348 PCa patients were recruited from two institutions for this study. The first institution contributed 284 patients, stratified and randomly divided into training and internal validation cohorts at a 7:3 ratio. The remaining 64 patients were sourced from the second institution and comprised the external validation cohort. Radiomics features were extracted from axial T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI) tumor regions. We developed the radiomics prediction model for BM in the training cohort and validated it in the internal and external validation cohorts. As a benchmark, we trained the logistic regression model with lasso feature reduction (LFR-LRM) in the training cohort and further compared it with Naive Bayes, eXtreme Gradient Boosting (XGboost), Random Forest (RF), GBDT, SVM, Adaboost, and KNN algorithms and validated in both the internal and external cohorts. The performance of several predictive models was assessed by receiver operating characteristic (ROC).

RESULTS

The LFR-LRM model achieved an area under the receiver operating characteristic curve (AUC) of 0.89 (95% CI: 0.822-0.974) and an accuracy of 0.828 (95% CI: 0.713-0.911). The AUC and accuracy in external validation were 0.866 (95% CI: 0.784-0.948) and 0.769 (95% CI: 0.648-0.864), respectively. The RF and XGBoost models outperformed the LFR-LRM, with AUCs of 0.907 (95% CI: 0.863-0.949) and 0.928 (95% CI: 0.882-0.974) and accuracies of 0.831 (95% CI: 0.727-0.907) and 0.884 (95% CI: 0.792-0.946). External validation for these models yielded AUCs and accuracies of 0.911 (95% CI: 0.861-0.966), 0.921 (95% CI: 0.889-0.953), and 0.846 (95% CI: 0.735-0.923) and 0.876 (95% CI: 0.771-0.945), respectively.

CONCLUSIONS

The XGboost machine learning model is more accurate than LFR-LRM for predicting BM in patients with newly confirmed PCa.

Accurate detection of acute sleep deprivation using a metabolomic biomarker-A machine learning approach

Sleep deprivation enhances risk for serious injury and fatality on the roads and in workplaces. To facilitate future management of these risks through advanced detection, we developed and validated a metabolomic biomarker of sleep deprivation in healthy, young participants, across three experiments. Bi-hourly plasma samples from 2 × 40-hour extended wake protocols (for train/test models) and 1 × 40-hour protocol with an 8-hour overnight sleep interval were analyzed by untargeted liquid chromatography-mass spectrometry. Using a knowledge-based machine learning approach, five consistently important variables were used to build predictive models. Sleep deprivation (24 to 38 hours awake) was predicted accurately in classification models [versus well-rested (0 to 16 hours)] (accuracy = 94.7%/AUC 99.2%, 79.3%/AUC 89.1%) and to a lesser extent in regression (R2 = 86.1 and 47.8%) models for within- and between-participant models, respectively. Metabolites were identified for replicability/future deployment. This approach for detecting acute sleep deprivation offers potential to reduce accidents through "fitness for duty" or "post-accident analysis" assessments.

Identification of TMZ resistance-associated histone post-translational modifications in glioblastoma using multi-omics data

BACKGROUD

Glioblastoma multiforme (GBM) is among the most aggressive cancers, with current treatments limited in efficacy. A significant hurdle in the treatment of GBM is the resistance to the chemotherapeutic agent temozolomide (TMZ). The methylation status of the MGMT promoter has been implicated as a critical biomarker of response to TMZ.

METHODS

To explore the mechanisms underlying resistance, we developed two TMZ-resistant GBM cell lines through a gradual increase in TMZ exposure. Transcriptome sequencing of TMZ-resistant cell lines revealed that alterations in histone post-translational modifications might be instrumental in conferring TMZ resistance. Subsequently, multi-omics analysis suggests a strong association between histone H3 lysine 9 acetylation (H3K9ac) levels and TMZ resistance.

RESULTS

We observed a significant correlation between the expression of H3K9ac and MGMT, particularly in the unmethylated MGMT promoter samples. More importantly, our findings suggest that H3K9ac may enhance MGMT transcription by facilitating the recruitment of the SP1 transcription factor to the MGMT transcription factor binding site. Additionally, by analyzing single-cell transcriptomics data from matched primary and recurrent GBM tumors treated with TMZ, we modeled the molecular shifts occurring upon tumor recurrence. We also noted a reduction in tumor stem cell characteristics, accompanied by an increase in H3K9ac, SP1, and MGMT levels, underscoring the potential role of H3K9ac in tumor relapse following TMZ therapy.

CONCLUSIONS

The increase in H3K9ac appears to enhance the recruitment of the transcription factor SP1 to its binding sites within the MGMT locus, consequently upregulating MGMT expression and driving TMZ resistance in GBM.

A Comparison of Machine Learning Approaches for predicting Hepatotoxicity potential using Chemical Structure and Targeted Transcriptomic Data

Animal toxicity testing is time and resource intensive, making it difficult to keep pace with the number of substances requiring assessment. Machine learning (ML) models that use chemical structure information and high-throughput experimental data can be helpful in predicting potential toxicity . However, much of the toxicity data used to train ML models is biased with an unequal balance of positives and negatives primarily since substances selected for in vivo testing are expected to elicit some toxicity effect. To investigate the impact this bias had on predictive performance, various sampling approaches were used to balance in vivo toxicity data as part of a supervised ML workflow to predict hepatotoxicity outcomes from chemical structure and/or targeted transcriptomic data. From the chronic, subchronic, developmental, multigenerational reproductive, and subacute repeat-dose testing toxicity outcomes with a minimum of 50 positive and 50 negative substances, 18 different study-toxicity outcome combinations were evaluated in up to 7 ML models. These included Artificial Neural Networks, Random Forests, Bernouilli Naïve Bayes, Gradient Boosting, and Support Vector classification algorithms which were compared with a local approach, Generalised Read-Across (GenRA), a similarity-weighted k-Nearest Neighbour (k-NN) method. The mean CV F1 performance for unbalanced data across all classifiers and descriptors for chronic liver effects was 0.735 (0.0395 SD). Mean CV F1 performance dropped to 0.639 (0.073 SD) with over-sampling approaches though the poorer performance of KNN approaches in some cases contributed to the observed decrease (mean CV F1 performance excluding KNN was 0.697 (0.072 SD)). With under-sampling approaches, the mean CV F1 was 0.523 (0.083 SD). For developmental liver effects, the mean CV F1 performance was much lower with 0.089 (0.111 SD) for unbalanced approaches and 0.149 (0.084 SD) for under-sampling. Over-sampling approaches led to an increase in mean CV F1 performance (0.234, (0.107 SD)) for developmental liver toxicity. Model performance was found to be dependent on dataset, model type, balancing approach and feature selection. Accordingly tailoring ML workflows for predicting toxicity should consider class imbalance and rely on simpler classifiers first.

Dosiomics and radiomics-based prediction of pneumonitis after radiotherapy and immune checkpoint inhibition: The relevance of fractionation

OBJECTIVES

Post-therapy pneumonitis (PTP) is a relevant side effect of thoracic radiotherapy and immunotherapy with checkpoint inhibitors (ICI). The influence of the combination of both, including dose fractionation schemes on PTP development is still unclear. This study aims to improve the PTP risk estimation after radio(chemo)therapy (R(C)T) for lung cancer with and without ICI by investigation of the impact of dose fractionation on machine learning (ML)-based prediction.

MATERIALS AND METHODS

Data from 100 patients who received fractionated R(C)T were collected. 39 patients received additional ICI therapy. Computed Tomography (CT), RT segmentation and dose data were extracted and physical doses were converted to 2-Gy equivalent doses (EQD2) to account for different fractionation schemes. Features were reduced using Pearson intercorrelation and the Boruta algorithm within 1000-fold bootstrapping. Six single (clinics, Dose Volume Histogram (DVH), ICI, chemotherapy, radiomics, dosiomics) and four combined models (radiomics + dosiomics, radiomics + DVH + Clinics, dosiomics + DVH + Clinics, radiomics + dosiomics + DVH + Clinics) were trained to predict PTP. Dose-based models were tested using physical dose and EQD2. Four ML-algorithms (random forest (rf), logistic elastic net regression, support vector machine, logitBoost) were trained and tested using 5-fold nested cross validation and Synthetic Minority Oversampling Technique (SMOTE) for resampling in R. Prediction was evaluated using the area under the receiver operating characteristic curve (AUC) on the test sets of the outer folds.

RESULTS

The combined model of all features using EQD2 surpassed all other models (AUC = 0.77, Confidence Interval CI 0.76-0.78). DVH, clinical data and ICI therapy had minor impact on PTP prediction with AUC values between 0.42 and 0.57. All EQD2-based models outperformed models based on physical dose.

CONCLUSIONS

Radiomics + dosiomics based ML models combined with clinical and dosimetric models were found to be suited best for PTP prediction after R(C)T and could improve pre-treatment decision making. Different RT dose fractionation schemes should be considered for dose-based ML approaches.

An immune-related gene prognostic prediction risk model for neoadjuvant chemoradiotherapy in rectal cancer using artificial intelligence

Background

This study aimed to establish and validate a prognostic model based on immune-related genes (IRGPM) for predicting disease-free survival (DFS) in patients with locally advanced rectal cancer (LARC) undergoing neoadjuvant chemoradiotherapy, and to elucidate the immune profiles associated with different prognostic outcomes.

Methods

Transcriptomic and clinical data were sourced from the Gene Expression Omnibus (GEO) database and the West China Hospital database. We focused on genes from the RNA immune-oncology panel. The elastic net approach was employed to pinpoint immune-related genes significantly impacting DFS. We developed the IRGPM for rectal cancer using the random forest technique. Based on the IRGPM, we calculated prognostic risk scores to categorize patients into high-risk and low-risk groups. Comparative analysis of immune characteristics between these groups was conducted.

Results

In this study, 407 LARC samples were analyzed. The elastic net identified a signature of 20 immune-related genes, forming the basis of the IRGPM. Kaplan-Meier survival analysis revealed a lower 5-year DFS in the high-risk group compared to the low-risk group. The receiver operating characteristic (ROC) curve affirmed the model's robust predictive capability. Validation of the model was performed in the GSE190826 cohort and our institution's cohort. Gene expression differences between high-risk and low-risk groups predominantly related to cytokine-cytokine receptor interactions. Notably, the low-risk group exhibited higher immune scores. Further analysis indicated a greater presence of activated B cells, activated CD8 T cells, central memory CD8 T cells, macrophages, T follicular helper cells, and type 2 helper cells in the low-risk group. Additionally, immune checkpoint analysis revealed elevated PDCD1 expression in the low-risk group.

Conclusions

The IRGPM, developed through random forest and elastic net methodologies, demonstrates potential in distinguishing DFS among LARC patients receiving standard treatment. Notably, the low-risk group, as defined by the IRGPM, showed enhanced activation of adaptive immune responses within the tumor microenvironment.

Immune Cell Infiltration Types as Biomarkers for the Recurrence Diagnosis and Prognosis of Bladder Cancer

This study aimed to investigate the role of infiltrating immune cell types in diagnosing and predicting bladder cancer recurrence. This study mainly applied some algorithms, including Estimate the Proportion of Immune and Cancer Cells (EPIC), support vector machine-recursive feature elimination (SVM-RFE), random forest out-of-bag (RF-OOB) and least absolute shrinkage and selection operator (LASSO)-Cox regression analysis. We found six immune infiltrating cell types significantly associated with recurrence prognosis and two independent clinical prognostic factors. Infiltrating immune cell types (IICTs) based on the prognostic immune risk score (pIRS) models may provide significant biomarkers for the diagnosis and prognostic prediction of bladder cancer recurrence.

Random Forest Modeling of Acute Toxicity in Anal Cancer: Effects of Peritoneal Cavity Contouring Approaches on Model Performance

PURPOSE

Our purpose was to analyze the effect on gastrointestinal (GI) toxicity models when their dose-volume metrics predictors are derived from segmentations of the peritoneal cavity after different contouring approaches.

METHODS AND MATERIALS

A random forest machine learning approach was used to predict acute grade ≥3 GI toxicity from dose-volume metrics and clinicopathologic factors for 246 patients (toxicity incidence = 9.5%) treated with definitive chemoradiation for squamous cell carcinoma of the anus. Three types of random forest models were constructed based on different bowel bag segmentation approaches: (1) physician-delineated after Radiation Therapy Oncology Group (RTOG) guidelines, (2) autosegmented by a deep learning model (nnU-Net) following RTOG guidelines, and (3) autosegmented but spanning the entire bowel space. Each model type was evaluated using repeated cross-validation (100 iterations; 50%/50% training/test split). The performance of the models was assessed using area under the precision-recall curve (AUPRC) and the receiver operating characteristic curve (AUROCC), as well as optimal F1 score.

RESULTS

When following RTOG guidelines, the models based on the nnU-Net auto segmentations (mean values: AUROCC, 0.71 ± 0.07; AUPRC, 0.42 ± 0.09; F1 score, 0.46 ± 0.08) significantly outperformed (P < .001) those based on the physician-delineated contours (mean values: AUROCC, 0.67 ± 0.07; AUPRC, 0.34 ± 0.08; F1 score, 0.36 ± 0.07). When spanning the entire bowel space, the performance of the autosegmentation models improved considerably (mean values: AUROCC, 0.87 ± 0.05; AUPRC, 0.70 ± 0.09; F1 score, 0.68 ± 0.09).

CONCLUSIONS

Random forest models were superior at predicting acute grade ≥3 GI toxicity when based on RTOG-defined bowel bag autosegmentations rather than physician-delineated contours. Models based on autosegmentations spanning the entire bowel space show further considerable improvement in model performance. The results of this study should be further validated using an external data set.

A critical moment in machine learning in medicine: on reproducible and interpretable learning

Over the past two decades, advances in computational power and data availability combined with increased accessibility to pre-trained models have led to an exponential rise in machine learning (ML) publications. While ML may have the potential to transform healthcare, this sharp increase in ML research output without focus on methodological rigor and standard reporting guidelines has fueled a reproducibility crisis. In addition, the rapidly growing complexity of these models compromises their interpretability, which currently impedes their successful and widespread clinical adoption. In medicine, where failure of such models may have severe implications for patients' health, the high requirements for accuracy, robustness, and interpretability confront ML researchers with a unique set of challenges. In this review, we discuss the semantics of reproducibility and interpretability, as well as related issues and challenges, and outline possible solutions to counteracting the "black box". To foster reproducibility, standard reporting guidelines need to be further developed and data or code sharing encouraged. Editors and reviewers may equally play a critical role by establishing high methodological standards and thus preventing the dissemination of low-quality ML publications. To foster interpretable learning, the use of simpler models more suitable for medical data can inform the clinician how results are generated based on input data. Model-agnostic explanation tools, sensitivity analysis, and hidden layer representations constitute further promising approaches to increase interpretability. Balancing model performance and interpretability are important to ensure clinical applicability. We have now reached a critical moment for ML in medicine, where addressing these issues and implementing appropriate solutions will be vital for the future evolution of the field.

Construction of a diagnostic model for osteoarthritis based on transcriptomic immune-related genes

Background

Osteoarthritis (OA) is a leading cause of disability globally, affecting over 500 million individuals worldwide. However, accurate and early diagnosis of OA is challenging to achieve. Immune-related genes play an essential role in OA development. Therefore, the objective of this study was to develop a diagnostic model for OA based on immune-related genes identified in synovial membrane.

Methods

The gene expression profile of OA were downloaded based on four datasets. The significantly differentially expressed genes (DEGs) between OA and control groups were selected. The differential immune cells were analyzed, followed by immune-related DEGs screening. WGCNA was used to screen module genes and these genes were further selected through optimization algorithm. Then, nomogram model was constructed. Chemical drug small molecule related to OA was predicted. Finally, expression levels of several key genes were validated by qRT-PCR through construction of OA rat models.

Results

The total 656 DEGs were obtained. Eight immune cells were significantly differential between two groups, and 317 immune-related DEGs were obtained. WGCNA identified three modules. The genes in modules were significantly involved in 15 pathways, involving in 65 genes. Then 12 DEGs were screened as the final optimal combination of DEGs, such as CEBPB, CXCL1, JUND, GABARAPL2 and PDGFC. The Nomogram model was also constructed. Furthermore, the chemical small molecules, such as acetaminophen, aspirin, and caffeine were predicted. The expression levels of CEBPB, CXCL1, GABARAPL2 and PDGFC were validated in OA rat models.

Conclusion

A diagnostic model based on twelve immune related genes was constructed. These model genes, such as CEBPB, CXCL1, GABARAPL2, and PDGFC, may serve as diagnostic biomarkers and immunotherapeutic targets.

Bioinformatics approach to identify the hub gene associated with COVID-19 and idiopathic pulmonary fibrosis

The coronavirus disease 2019 (COVID-19) has developed into a global health crisis. Pulmonary fibrosis, as one of the complications of SARS-CoV-2 infection, deserves attention. As COVID-19 is a new clinical entity that is constantly evolving, and many aspects of disease are remain unknown. The datasets of COVID-19 and idiopathic pulmonary fibrosis were obtained from the Gene Expression Omnibus. The hub genes were screened out using the Random Forest (RF) algorithm depending on the severity of patients with COVID-19. A risk prediction model was developed to assess the prognosis of patients infected with SARS-CoV-2, which was evaluated by another dataset. Six genes (named NELL2, GPR183, S100A8, ALPL, CD177, and IL1R2) may be associated with the development of PF in patients with severe SARS-CoV-2 infection. S100A8 is thought to be an important target gene that is closely associated with COVID-19 and pulmonary fibrosis. Construction of a neural network model was successfully predicted the prognosis of patients with COVID-19. With the increasing availability of COVID-19 datasets, bioinformatic methods can provide possible predictive targets for the diagnosis, treatment, and prognosis of the disease and show intervention directions for the development of clinical drugs and vaccines.

Machine Learning and Radiomics of Bone Scintigraphy: Their Role in Predicting Recurrence of Localized or Locally Advanced Prostate Cancer

BACKGROUND

Machine-learning (ML) and radiomics features have been utilized for survival outcome analysis in various cancers. This study aims to investigate the application of ML based on patients' clinical features and radiomics features derived from bone scintigraphy (BS) and to evaluate recurrence-free survival in local or locally advanced prostate cancer (PCa) patients after the initial treatment.

METHODS

A total of 354 patients who met the eligibility criteria were analyzed and used to train the model. Clinical information and radiomics features of BS were obtained. Survival-related clinical features and radiomics features were included in the ML model training. Using the pyradiomics software, 128 radiomics features from each BS image's region of interest, validated by experts, were extracted. Four textural matrices were also calculated: GLCM, NGLDM, GLRLM, and GLSZM. Five training models (Logistic Regression, Naive Bayes, Random Forest, Support Vector Classification, and XGBoost) were applied using K-fold cross-validation. Recurrence was defined as either a rise in PSA levels, radiographic progression, or death. To assess the classifier's effectiveness, the ROC curve area and confusion matrix were employed.

RESULTS

Of the 354 patients, 101 patients were categorized into the recurrence group with more advanced disease status compared to the non-recurrence group. Key clinical features including tumor stage, radical prostatectomy, initial PSA, Gleason Score primary pattern, and radiotherapy were used for model training. Random Forest (RF) was the best-performing model, with a sensitivity of 0.81, specificity of 0.87, and accuracy of 0.85. The ROC curve analysis showed that predictions from RF outperformed predictions from other ML models with a final AUC of 0.94 and a p-value of <0.001. The other models had accuracy ranges from 0.52 to 0.78 and AUC ranges from 0.67 to 0.84.

CONCLUSIONS

The study showed that ML based on clinical features and radiomics features of BS improves the prediction of PCa recurrence after initial treatment. These findings highlight the added value of ML techniques for risk classification in PCa based on clinical features and radiomics features of BS.

Identification of a prognostic evaluator from glutamine metabolic heterogeneity studies within and between tissues in hepatocellular carcinoma

Background: The liver is the major metabolic organ of the human body, and abnormal metabolism is the main factor influencing hepatocellular carcinoma (HCC). This study was designed to determine the effect of glutamine metabolism on HCC heterogeneity and to develop a prognostic evaluator based on the heterogeneity study of glutamine metabolism within HCC tumors and between tissues. Methods: Single-cell transcriptome data were extracted from the GSE149614 dataset and processed using the Seurat package in R for quality control of these data. HCC subtypes in the Cancer Genome Atlas and the GSE14520 dataset were identified via consensus clustering based on glutamine family amino acid metabolism (GFAAM) process genes. The machine learning algorithms gradient boosting machine, support vector machine, random forest, eXtreme gradient boosting, decision trees, and least absolute shrinkage and selection operator were utilized to develop the prognosis model of differentially expressed genes among the molecular gene subtypes. Results: The samples in the GSE149614 dataset included 10 cell types, and there was no significant difference in the GFAAM pathway. HCC was classified into three molecular subtypes according to GFAAM process genes, showing molecular heterogeneity in prognosis, clinicopathological features, and immune cell infiltration. C1 showed the worst survival rate and the highest immune score and immune cell infiltration. A six-gene model for prognostic and immunotherapy responses was constructed among subtypes, and the calculated high-risk score was significantly correlated with poor prognosis, high immune abundance, and a low response rate of immunotherapy in HCC. Conclusion: Our discovery of GFAAM-associated marker genes may help to further decipher the role in HCC occurrence and progression. In particular, this six-gene prognostic model may serve as a predictor of treatment and prognosis in HCC patients.

Machine learning approach for the detection of vitamin D level: a comparative study

BACKGROUND

After the World Health Organization declared the COVID-19 pandemic, the role of Vitamin D has become even more critical for people worldwide. The most accurate way to define vitamin D level is 25-hydroxy vitamin D(25-OH-D) blood test. However, this blood test is not always feasible. Most data sets used in health science research usually contain highly correlated features, which is referred to as multicollinearity problem. This problem can lead to misleading results and overfitting problems in the ML training process. Therefore, the proposed study aims to determine a clinically acceptable ML model for the detection of the vitamin D status of the North Cyprus adult participants accurately, without the need to determine 25-OH-D level, taking into account the multicollinearity problem.

METHOD

The study was conducted with 481 observations who applied voluntarily to Internal Medicine Department at NEU Hospital. The classification performance of four conventional supervised ML models, namely, Ordinal logistic regression(OLR), Elastic-net ordinal regression(ENOR), Support Vector Machine(SVM), and Random Forest (RF) was compared. The comparative analysis is performed regarding the model's sensitivity to the participant's metabolic syndrome(MtS)'positive status, hyper-parameter tuning, sensitivities to the size of training data, and the classification performance of the models.

RESULTS

Due to the presence of multicollinearity, the findings showed that the performance of the SVM(RBF) is obviously negatively affected when the test is examined. Moreover, it can be obviously detected that RF is more robust than other models when the variations in the size of training data are examined. This experiment's result showed that the selected RF and ENOR showed better performances than the other two models when the size of training samples was reduced. Since the multicollinearity is more severe in the small samples, it can be concluded that RF and ENOR are not affected by the presence of the multicollinearity problem. The comparative analysis revealed that the RF classifier performed better and was more robust than the other proposed models in terms of accuracy (0.94), specificity (0.96), sensitivity or recall (0.94), precision (0.95), F1-score (0.95), and Cohen's kappa (0.90).

CONCLUSION

It is evident that the RF achieved better than the SVM(RBF), ENOR, and OLR. These comparison findings will be applied to develop a Vitamin D level intelligent detection system for being used in routine clinical, biochemical tests, and lifestyle characteristics of individuals to decrease the cost and time of vitamin D level detection.

Multitask Learning with Convolutional Neural Networks and Vision Transformers Can Improve Outcome Prediction for Head and Neck Cancer Patients

Neural-network-based outcome predictions may enable further treatment personalization of patients with head and neck cancer. The development of neural networks can prove challenging when a limited number of cases is available. Therefore, we investigated whether multitask learning strategies, implemented through the simultaneous optimization of two distinct outcome objectives (multi-outcome) and combined with a tumor segmentation task, can lead to improved performance of convolutional neural networks (CNNs) and vision transformers (ViTs). Model training was conducted on two distinct multicenter datasets for the endpoints loco-regional control (LRC) and progression-free survival (PFS), respectively. The first dataset consisted of pre-treatment computed tomography (CT) imaging for 290 patients and the second dataset contained combined positron emission tomography (PET)/CT data of 224 patients. Discriminative performance was assessed by the concordance index (C-index). Risk stratification was evaluated using log-rank tests. Across both datasets, CNN and ViT model ensembles achieved similar results. Multitask approaches showed favorable performance in most investigations. Multi-outcome CNN models trained with segmentation loss were identified as the optimal strategy across cohorts. On the PET/CT dataset, an ensemble of multi-outcome CNNs trained with segmentation loss achieved the best discrimination (C-index: 0.29, 95% confidence interval (CI): 0.22-0.36) and successfully stratified patients into groups with low and high risk of disease progression (p=0.003). On the CT dataset, ensembles of multi-outcome CNNs and of single-outcome ViTs trained with segmentation loss performed best (C-index: 0.26 and 0.26, CI: 0.18-0.34 and 0.18-0.35, respectively), both with significant risk stratification for LRC in independent validation (p=0.002 and p=0.011). Further validation of the developed multitask-learning models is planned based on a prospective validation study, which has recently completed recruitment.

CT-based deep multi-label learning prediction model for outcome in patients with oropharyngeal squamous cell carcinoma

BACKGROUND

Personalized treatment is increasingly required for oropharyngeal squamous cell carcinoma (OPSCC) patients due to emerging new cancer subtypes and treatment options. Outcome prediction model can help identify low or high-risk patients who may be suitable to receive de-escalation or intensified treatment approaches.

PURPOSE

To develop a deep learning (DL)-based model for predicting multiple and associated efficacy endpoints in OPSCC patients based on computed tomography (CT).

METHODS

Two patient cohorts were used in this study: a development cohort consisting of 524 OPSCC patients (70% for training and 30% for independent testing) and an external test cohort of 396 patients. Pre-treatment CT-scans with the gross primary tumor volume contours (GTVt) and clinical parameters were available to predict endpoints, including 2-year local control (LC), regional control (RC), locoregional control (LRC), distant metastasis-free survival (DMFS), disease-specific survival (DSS), overall survival (OS), and disease-free survival (DFS). We proposed DL outcome prediction models with the multi-label learning (MLL) strategy that integrates the associations of different endpoints based on clinical factors and CT-scans.

RESULTS

The multi-label learning models outperformed the models that were developed based on a single endpoint for all endpoints especially with high AUCs ≥ 0.80 for 2-year RC, DMFS, DSS, OS, and DFS in the internal independent test set and for all endpoints except 2-year LRC in the external test set. Furthermore, with the models developed, patients could be stratified into high and low-risk groups that were significantly different for all endpoints in the internal test set and for all endpoints except DMFS in the external test set.

CONCLUSION

MLL models demonstrated better discriminative ability for all 2-year efficacy endpoints than single outcome models in the internal test and for all endpoints except LRC in the external set.

Towards real-time EPID-based 3D in vivo dosimetry for IMRT with Deep Neural Networks: A feasibility study

We investigate the potential of the Deep Dose Estimate (DDE) neural network to predict 3D dose distributions inside patients with Monte Carlo (MC) accuracy, based on transmitted EPID signals and patient CTs. The network was trained using as input patient CTs and first-order dose approximations (FOD). Accurate dose distributions (ADD) simulated with MC were given as training targets. 83 pelvic CTs were used to simulate ADDs and respective EPID signals for subfields of prostate IMRT plans (gantry at 0∘). FODs were produced as backprojections from the EPID signals. 581 ADD-FOD sets were produced and divided into training and test sets. An additional dataset simulated with gantry at 90∘ (lateral set) was used for evaluating the performance of the DDE at different beam directions. The quality of the FODs and DDE-predicted dose distributions (DDEP) with respect to ADDs, from the test and lateral sets, was evaluated with gamma analysis (3%,2 mm). The passing rates between FODs and ADDs were as low as 46%, while for DDEPs the passing rates were above 97% for the test set. Meaningful improvements were also observed for the lateral set. The high passing rates for DDEPs indicate that the DDE is able to convert FODs into ADDs. Moreover, the trained DDE predicts the dose inside a patient CT within 0.6 s/subfield (GPU), in contrast to 14 h needed for MC (CPU-cluster). 3D in vivo dose distributions due to clinical patient irradiation can be obtained within seconds, with MC-like accuracy, potentially paving the way towards real-time EPID-based in vivo dosimetry.

EviGUIDE - a tool for evidence-based decision making in image-guided adaptive brachytherapy for cervical cancer

PURPOSE

To develop a novel decision-support system for radiation oncology that incorporates clinical, treatment and outcome data, as well as outcome models from a large clinical trial on magnetic resonance image-guided adaptive brachytherapy (MR-IGABT) for locally advanced cervical cancer (LACC).

METHODS

A system, called EviGUIDE, was developed that combines dosimetric information from the treatment planning system, patient and treatment characteristics, and established tumor control probability (TCP), and normal tissue complication probability (NTCP) models, to predict clinical outcome of radiotherapy treatment of LACC. Six Cox Proportional Hazards models based on data from 1341 patients of the EMBRACE-I study have been integrated. One TCP model for local tumor control, and five NTCP models for OAR morbidities.

RESULTS

EviGUIDE incorporates TCP-NTCP graphs to help users visualize the clinical impact of different treatment plans and provides feedback on achievable doses based on a large reference population. It enables holistic assessment of the interplay between multiple clinical endpoints and tumour and treatment variables. Retrospective analysis of 45 patients treated with MR-IGABT showed that there exists a sub-cohort of patients (20%) with increased risk factors, that could greatly benefit from the quantitative and visual feedback.

CONCLUSION

A novel digital concept was developed that can enhance clinical decision- making and facilitate personalized treatment. It serves as a proof of concept for a new generation of decision support systems in radiation oncology, which incorporate outcome models and high-quality reference data, and aids the dissemination of evidence-based knowledge about optimal treatment and serve as a blueprint for other sites in radiation oncology.

Bibliometric analysis of the global scientific production on machine learning applied to different cancer types

Cancer disease is one of the main causes of death in the world, with million annual cases in the last decades. The need to find a cure has stimulated the search for efficient treatments and diagnostic procedures. One of the most promising tools that has emerged against cancer in recent years is machine learning (ML), which has raised a huge number of scientific papers published in a relatively short period of time. The present study analyzes global scientific production on ML applied to the most relevant cancer types through various bibliometric indicators. We find that over 30,000 studies have been published so far and observe that cancers with the highest number of published studies using ML (breast, lung, and colon cancer) are those with the highest incidence, being the USA and China the main scientific producers on the subject. Interestingly, the role of China and Japan in stomach cancer is correlated with the number of cases of this cancer type in Asia (78% of the worldwide cases). Knowing the countries and institutions that most study each area can be of great help for improving international collaborations between research groups and countries. Our analysis shows that medical and computer science journals lead the number of publications on the subject and could be useful for researchers in the field. Finally, keyword co-occurrence analysis suggests that ML-cancer research trends are focused not only on the use of ML as an effective diagnostic method, but also for the improvement of radiotherapy- and chemotherapy-based treatments.

Construction of a Matrix Cancer-Associated Fibroblast Signature Gene-Based Risk Prognostic Signature for Directing Immunotherapy in Patients with Breast Cancer Using Single-Cell Analysis and Machine Learning

Cancer-associated fibroblasts (CAFs) are heterogeneous constituents of the tumor microenvironment involved in the tumorigenesis, progression, and therapeutic responses of tumors. This study identified four distinct CAF subtypes of breast cancer (BRCA) using single-cell RNA sequencing (RNA-seq) data. Of these, matrix CAFs (mCAFs) were significantly associated with tumor matrix remodeling and strongly correlated with the transforming growth factor (TGF)-β signaling pathway. Consensus clustering of The Cancer Genome Atlas (TCGA) BRCA dataset using mCAF single-cell characteristic gene signatures segregated samples into high-fibrotic and low-fibrotic groups. Patients in the high-fibrotic group exhibited a significantly poor prognosis. A weighted gene co-expression network analysis and univariate Cox analysis of bulk RNA-seq data revealed 17 differential genes with prognostic values. The mCAF risk prognosis signature (mRPS) was developed using 10 machine learning algorithms. The clinical outcome predictive accuracy of the mRPS was higher than that of the conventional TNM staging system. mRPS was correlated with the infiltration level of anti-tumor effector immune cells. Based on consensus prognostic genes, BRCA samples were classified into the following two subtypes using six machine learning algorithms (accuracy > 90%): interferon (IFN)-γ-dominant (immune C2) and TGF-β-dominant (immune C6) subtypes. Patients with mRPS downregulation were associated with improved prognosis, suggesting that they can potentially benefit from immunotherapy. Thus, the mRPS model can stably predict BRCA prognosis, reflect the local immune status of the tumor, and aid clinical decisions on tumor immunotherapy.

Construction and Validation of a Reliable Disulfidptosis-Related LncRNAs Signature of the Subtype, Prognostic, and Immune Landscape in Colon Cancer

Disulfidptosis, a novel form of regulated cell death (RCD) associated with metabolism, represents a promising intervention target in cancer therapy. While abnormal lncRNA expression is associated with colon cancer development, the prognostic potential and biological characteristics of disulfidptosis-related lncRNAs (DRLs) remain unclear. Consequently, the research aimed to discover a novel indication of DRLs with significant prognostic implications, and to investigate their possible molecular role in the advancement of colon cancer. Here, we acquired RNA-seq data, pertinent clinical data, and genomic mutations of colon adenocarcinoma (COAD) from the TCGA database, and then DRLs were determined through Pearson correlation analysis. A total of 434 COAD patients were divided in to three subgroups through clustering analysis based on DRLs. By utilizing univariate Cox regression, the least absolute shrinkage and selection operator (LASSO) algorithm, and multivariate Cox regression analysis, we ultimately created a prognostic model consisting of four DRLs (AC007728.3, AP003555.1, ATP2B1.AS1, and NSMCE1.DT), and an external database was used to validate the prognostic features of the risk model. According to the Kaplan-Meier curve analysis, patients in the low-risk group exhibited a considerably superior survival time in comparison to those in the high-risk group. Enrichment analysis revealed a significant association between metabolic processes and the genes that were differentially expressed in the high- and low-risk groups. Additionally, significant differences in the tumor immune microenvironment landscape were observed, specifically pertaining to immune cells, function, and checkpoints. High-risk patients exhibited a low likelihood of immune evasion, as indicated by the Tumor Immune Dysfunction and Exclusion (TIDE) analysis. Patients who exhibit both a high risk and high Tumor Mutational Burden (TMB) experience the least amount of time for survival, whereas those belonging to the low-risk and low-TMB category demonstrate the most favorable prognosis. In addition, the risk groups determined by the 4-DRLs signature displayed distinct drug sensitivities. Finally, we confirmed the levels of expression for four DRLs through rt-qPCR in both tissue samples from colon cancer patients and cell lines. Taken together, the first 4-DRLs-based signature we proposed may serve for a hopeful instrument for forecasting the prognosis, immune landscape, and therapeutic responses in colon cancer patients, thereby facilitating optimal clinical decision-making.

Expression patterns and immunological characterization of PANoptosis -related genes in gastric cancer

Background

Accumulative studies have demonstrated the close relationship between tumor immunity and pyroptosis, apoptosis, and necroptosis. However, the role of PANoptosis in gastric cancer (GC) is yet to be fully understood.

Methods

This research attempted to identify the expression patterns of PANoptosis regulators and the immune landscape in GC by integrating the GSE54129 and GSE65801 datasets. We analyzed GC specimens and established molecular clusters associated with PANoptosis-related genes (PRGs) and corresponding immune characteristics. The differentially expressed genes were determined with the WGCNA method. Afterward, we employed four machine learning algorithms (Random Forest, Support Vector Machine, Generalized linear Model, and eXtreme Gradient Boosting) to select the optimal model, which was validated using nomogram, calibration curve, decision curve analysis (DCA), and two validation cohorts. Additionally, this study discussed the relationship between infiltrating immune cells and variables in the selected model.

Results

This study identified dysregulated PRGs and differential immune activities between GC and normal samples, and further identified two PANoptosis-related molecular clusters in GC. These clusters demonstrated remarkable immunological heterogeneity, with Cluster1 exhibiting abundant immune infiltration. The Support Vector Machine signature was found to have the best discriminative ability, and a 5-gene-based SVM signature was established. This model showed excellent performance in the external validation cohorts, and the nomogram, calibration curve, and DCA indicated its reliability in predicting GC patterns. Further analysis confirmed that the 5 selected variables were remarkably related to infiltrating immune cells and immune-related pathways.

Conclusion

Taken together, this work demonstrates that the PANoptosis pattern has the potential as a stratification tool for patient risk assessment and a reflection of the immune microenvironment in GC.

Mean Heart Dose Prediction Using Parameters of Single-Slice Computed Tomography and Body Mass Index: Machine Learning Approach for Radiotherapy of Left-Sided Breast Cancer of Asian Patients

Deep inspiration breath-hold (DIBH) is an excellent technique to reduce the incidental radiation received by the heart during radiotherapy in patients with breast cancer. However, DIBH is costly and time-consuming for patients and radiotherapy staff. In Asian countries, the use of DIBH is restricted due to the limited number of patients with a high mean heart dose (MHD) and the shortage of radiotherapy personnel and equipment compared to that in the USA. This study aimed to develop, evaluate, and compare the performance of ten machine learning algorithms for predicting MHD using a patient's body mass index and single-slice CT parameters to identify patients who may not require DIBH. Machine learning models were built and tested using a dataset containing 207 patients with left-sided breast cancer who were treated with field-in-field radiotherapy with free breathing. The average MHD was 251 cGy. Stratified repeated four-fold cross-validation was used to build models using 165 training data. The models were compared internally using their average performance metrics: F2 score, AUC, recall, accuracy, Cohen's kappa, and Matthews correlation coefficient. The final performance evaluation for each model was further externally analyzed using 42 unseen test data. The performance of each model was evaluated as a binary classifier by setting the cut-off value of MHD ≥ 300 cGy. The deep neural network (DNN) achieved the highest F2 score (78.9%). Most models successfully classified all patients with high MHD as true positive. This study indicates that the ten models, especially the DNN, might have the potential to identify patients who may not require DIBH.

Machine learning models-based on integration of next-generation sequencing testing and tumor cell sizes improve subtype classification of mature B-cell neoplasms

Background

Next-generation sequencing (NGS) panels for mature B-cell neoplasms (MBNs) are widely applied clinically but have yet to be routinely used in a manner that is suitable for subtype differential diagnosis. This study retrospectively investigated newly diagnosed cases of MBNs from our laboratory to investigate mutation landscapes in Chinese patients with MBNs and to combine mutational information and machine learning (ML) into clinical applications for MBNs, especially for subtype classification.

Methods

Samples from the Catalogue Of Somatic Mutations In Cancer (COSMIC) database were collected for ML model construction and cases from our laboratory were used for ML model validation. Five repeats of 10-fold cross-validation Random Forest algorithm was used for ML model construction. Mutation detection was performed by NGS and tumor cell size was confirmed by cell morphology and/or flow cytometry in our laboratory.

Results

Totally 849 newly diagnosed MBN cases from our laboratory were retrospectively identified and included in mutational landscape analyses. Patterns of gene mutations in a variety of MBN subtypes were found, important to investigate tumorigenesis in MBNs. A long list of novel mutations was revealed, valuable to both functional studies and clinical applications. By combining gene mutation information revealed by NGS and ML, we established ML models that provide valuable information for MBN subtype classification. In total, 8895 cases of 8 subtypes of MBNs in the COSMIC database were collected and utilized for ML model construction, and the models were validated on the 849 MBN cases from our laboratory. A series of ML models was constructed in this study, and the most efficient model, with an accuracy of 0.87, was based on integration of NGS testing and tumor cell sizes.

Conclusions

The ML models were of great significance in the differential diagnosis of all cases and different MBN subtypes. Additionally, using NGS results to assist in subtype classification of MBNs by method of ML has positive clinical potential.

Development of a novel immune infiltration-related diagnostic model for Alzheimer's disease using bioinformatic strategies

Background

The pathogenesis of Alzheimer's disease (AD) is complex and multi-factorial. Increasing evidence has shown the important role of immune infiltration in AD. Thus the current study was designed to identify immune infiltration-related genes and to explore their diagnostic value in AD.

Methods

The expression data of AD patients were downloaded from the GEO database. The limma R package identified differentially expressed genes (DEGs) between AD and controls. The CIBERSORT algorithm identified differentially infiltrated immune cells (DIICs) between AD and controls. DIIC-correlated DEGs were obtained by Pearson correlation analysis. WGCNA was employed to identify DIIC-related modules. Next, LASSO, RFE, and RF machine learning methods were applied to screen robust DIIC-related gene signatures in AD, followed by the construction and validation of a diagnostic nomogram. Detection of the expression of related genes in the peripheral blood of Alzheimer's disease and healthy volunteers by RT-PCR. In addition, the CTD database predicted chemicals targeting DIIC-related gene signatures in the treatment of AD.

Results

NK cells, M0 macrophages, activated myeloid dendritic cells, resting mast cells, CD8+ T cells, resting memory CD4+ T cells, gamma delta T cells, and M2 macrophages were differentially infiltrated between AD and controls. Pearson analysis identified a total of 277 DIIC-correlated DEGs between AD and controls. Thereafter, 177 DIIC-related genes were further obtained by WGCNA analysis. By LASSO, RFE and RF algorithms, CMTM2, DDIT4, LDHB, NDUFA1, NDUFB2, NDUFS5, RPL17, RPL21, RPL26 and NDUFAF2 were identified as robust gene signature in AD. The results of RT-PCR detection of peripheral blood samples from Alzheimer's disease and healthy volunteers showed that the expression trend of ten genes screened was consistent with the detection results; among them, the expression levels of CMTM2, DDIT4, LDHB, NDUFS5, and RPL21 are significantly different among groups. Thus, a diagnostic nomogram based on a DIIC-related signature was constructed and validated. Moreover, candidate chemicals targeting those biomarkers in the treatment of AD, such as 4-hydroxy-2-nonenal, rosiglitazone, and resveratrol, were identified in the CTD database.

Conclusion

For the first time, we identified 10 immune infiltration-related biomarkers in AD, which may be helpful for the diagnosis of AD and provide guidance in the treatment of AD.

Biomarker Discovery in Rare Malignancies: Development of a miRNA Signature for RDEB-cSCC

Machine learning has been proven to be a powerful tool in the identification of diagnostic tumor biomarkers but is often impeded in rare cancers due to small patient numbers. In patients suffering from recessive dystrophic epidermolysis bullosa (RDEB), early-in-life development of particularly aggressive cutaneous squamous-cell carcinomas (cSCCs) represents a major threat and timely detection is crucial to facilitate prompt tumor excision. As miRNAs have been shown to hold great potential as liquid biopsy markers, we characterized miRNA signatures derived from cultured primary cells specific for the potential detection of tumors in RDEB patients. To address the limitation in RDEB-sample accessibility, we analyzed the similarity of RDEB miRNA profiles with other tumor entities derived from the Cancer Genome Atlas (TCGA) repository. Due to the similarity in miRNA expression with RDEB-SCC, we used HN-SCC data to train a tumor prediction model. Three models with varying complexity using 33, 10 and 3 miRNAs were derived from the elastic net logistic regression model. The predictive performance of all three models was determined on an independent HN-SCC test dataset (AUC-ROC: 100%, 83% and 96%), as well as on cell-based RDEB miRNA-Seq data (AUC-ROC: 100%, 100% and 91%). In addition, the ability of the models to predict tumor samples based on RDEB exosomes (AUC-ROC: 100%, 93% and 100%) demonstrated the potential feasibility in a clinical setting. Our results support the feasibility of this approach to identify a diagnostic miRNA signature, by exploiting publicly available data and will lay the base for an improvement of early RDEB-SCC detection.

Multilingual RECIST classification of radiology reports using supervised learning

Objectives

The objective of this study is the exploration of Artificial Intelligence and Natural Language Processing techniques to support the automatic assignment of the four Response Evaluation Criteria in Solid Tumors (RECIST) scales based on radiology reports. We also aim at evaluating how languages and institutional specificities of Swiss teaching hospitals are likely to affect the quality of the classification in French and German languages.

Methods

In our approach, 7 machine learning methods were evaluated to establish a strong baseline. Then, robust models were built, fine-tuned according to the language (French and German), and compared with the expert annotation.

Results

The best strategies yield average F1-scores of 90% and 86% respectively for the 2-classes (Progressive/Non-progressive) and the 4-classes (Progressive Disease, Stable Disease, Partial Response, Complete Response) RECIST classification tasks.

Conclusions

These results are competitive with the manual labeling as measured by Matthew's correlation coefficient and Cohen's Kappa (79% and 76%). On this basis, we confirm the capacity of specific models to generalize on new unseen data and we assess the impact of using Pre-trained Language Models (PLMs) on the accuracy of the classifiers.

Integrated analysis of multiple transcriptomic data identifies ST8SIA6‑AS1 and LINC01093 as potential biomarkers in HBV‑associated liver cancer

The mechanisms of long-non-coding RNAs (lncRNAs) in hepatitis B virus (HBV) infection-associated liver cancer remain largely unclear. Therefore, the aim of the present study was to investigate the regulatory mechanisms of lncRNAs in this disease. HBV-liver cancer related transcriptome expression profile data (GSE121248 and GSE55092) from the Gene Expression Omnibus database and survival prognosis information from The Cancer Genome Atlas (TCGA) database were obtained for analysis. The limma package was used to identify the overlapped differentially expressed RNAs (DERs), including DElncRNAs and DEmRNAs, in the GSE121248 and GSE55092 datasets. The screened optimized lncRNA signatures were used to develop a nomogram model based on the GSE121248 dataset, which was validated using the GSE55092 and TCGA datasets. A competitive endogenous RNA (ceRNA) network was constructed based on the screened prognosis-associated lncRNA signatures from TCGA dataset. In addition, the levels of specific lncRNAs were evaluated in HBV-infected human liver cancer tissues and cells, and Cell Counting Kit-8, ELISA and Transwell assays were performed to evaluate the effects of the lncRNAs in HBV-expressing liver cancer cells. A total of 535 overlapped DERs, including 30 DElncRNAs and 505 DEmRNAs, were identified in the GSE121248 and GSE55092 datasets. An optimized DElncRNA signature comprising 10 lncRNAs was used to establish a nomogram. ST8SIA6-AS1 and LINC01093 were identified as lncRNAs associated with HBV-liver cancer prognosis in TCGA dataset, and were applied to construct a ceRNA network. Reverse transcription-quantitative PCR analysis showed that ST8SIA6-AS1 was upregulated and LINC01093 was downregulated in HBV-infected human liver cancer tissues and HBV-expressing liver cancer cells compared with non-HBV-infected controls. ST8SIA6-AS1 knockdown and LINC01093 overexpression independently reduced the number of copies of HBV DNA, the levels of hepatitis B surface antigen and hepatitis B e antigen, as well as cell proliferation, migration and invasion. In summary, the present study identified ST8SIA6-AS1 and LINC01093 as two potential biomarkers that may be effective therapeutic targets for HBV-associated liver cancer.

Metabolic-related gene signatures for survival prediction and immune cell subtypes associated with prognosis in intrahepatic cholangiocarcinoma

OBJECTIVES

Our study aimed to reveal the metabolic-related gene signatures for survival prediction and immune cell subtypes associated with IHCC prognosis.

METHODS

Differentially expressed metabolic genes were identified between survival group and dead group which were divided according to survival at discharge. Recursive feature elimination (RFE) and randomForest (RF) algorithms were applied to optimize the combination of feature metabolic genes, which were used to generate SVM classifier. Performance of SVM classifier was evaluated by receiver operating characteristic (ROC) curves. Gene set enrichment analysis (GSEA) was conducted to uncover the activated pathways in high risk group, and differences in immune cell distributions were revealed.

RESULTS

There were 143 differentially expressed metabolic gens. RFE and RF identified 21 overlapping differentially expressed metabolic genes, and the constructed SVM classifier had excellent accuracy in training and validation dataset. RS survival prediction model was consisted of 10 metabolic genes. RS model had reliable predictive capability in the training and validation dataset. GSEA revealed 15 significant KEGG pathways that were relatively activated in the high risk group. High risk group had obviously lower counts of B cell naive and T cell CD4+ memory resting, while higher counts of B cell plasma and macrophage M2.

CONCLUSION

Prognostic prediction model of 10 metabolic genes could accurately predict the prognosis of IHCC patients.

Automatic reconstruction of interstitial needles using CT images in post-operative cervical cancer brachytherapy based on deep learning

Purpose

The purpose of this study was to investigate the precision of deep learning (DL)-based auto-reconstruction in localizing interstitial needles in post-operative cervical cancer brachytherapy (BT) using three-dimensional (3D) computed tomography (CT) images.

Material and methods

A convolutional neural network (CNN) was developed and presented for automatic reconstruction of interstitial needles. Data of 70 post-operative cervical cancer patients who received CT-based BT were used to train and test this DL model. All patients were treated with three metallic needles. Dice similarity coefficient (DSC), 95% Hausdorff distance (95% HD), and Jaccard coefficient (JC) were applied to evaluate the geometric accuracy of auto-reconstruction for each needle. Dose-volume indexes (DVI) between manual and automatic methods were used to analyze the dosimetric difference. Correlation between geometric metrics and dosimetric difference was evaluated using Spearman correlation analysis.

Results

The mean DSC values of DL-based model were 0.88, 0.89, and 0.90 for three metallic needles. Wilcoxon signed-rank test indicated no significant dosimetric differences in all BT planning structures between manual and automatic reconstruction methods (p > 0.05). Spearman correlation analysis demonstrated weak link between geometric metrics and dosimetry differences.

Conclusions

DL-based reconstruction method can be used to precisely localize the interstitial needles in 3D-CT images. The proposed automatic approach could improve the consistency of treatment planning for post-operative cervical cancer brachytherapy.

Dosiomics and radiomics to predict pneumonitis after thoracic stereotactic body radiotherapy and immune checkpoint inhibition

Introduction

Pneumonitis is a relevant side effect after radiotherapy (RT) and immunotherapy with checkpoint inhibitors (ICIs). Since the effect is radiation dose dependent, the risk increases for high fractional doses as applied for stereotactic body radiation therapy (SBRT) and might even be enhanced for the combination of SBRT with ICI therapy. Hence, patient individual pre-treatment prediction of post-treatment pneumonitis (PTP) might be able to support clinical decision making. Dosimetric factors, however, use limited information and, thus, cannot exploit the full potential of pneumonitis prediction.

Methods

We investigated dosiomics and radiomics model based approaches for PTP prediction after thoracic SBRT with and without ICI therapy. To overcome potential influences of different fractionation schemes, we converted physical doses to 2 Gy equivalent doses (EQD2) and compared both results. In total, four single feature models (dosiomics, radiomics, dosimetric, clinical factors) were tested and five combinations of those (dosimetric+clinical factors, dosiomics+radiomics, dosiomics+dosimetric+clinical factors, radiomics+dosimetric+clinical factors, radiomics+dosiomics+dosimetric+clinical factors). After feature extraction, a feature reduction was performed using pearson intercorrelation coefficient and the Boruta algorithm within 1000-fold bootstrapping runs. Four different machine learning models and the combination of those were trained and tested within 100 iterations of 5-fold nested cross validation.

Results

Results were analysed using the area under the receiver operating characteristic curve (AUC). We found the combination of dosiomics and radiomics features to outperform all other models with AUCradiomics+dosiomics, D = 0.79 (95% confidence interval 0.78-0.80) and AUCradiomics+dosiomics, EQD2 = 0.77 (0.76-0.78) for physical dose and EQD2, respectively. ICI therapy did not impact the prediction result (AUC ≤ 0.5). Clinical and dosimetric features for the total lung did not improve the prediction outcome.

Conclusion

Our results suggest that combined dosiomics and radiomics analysis can improve PTP prediction in patients treated with lung SBRT. We conclude that pre-treatment prediction could support clinical decision making on an individual patient basis with or without ICI therapy.

External validation and comparison of MR-based radiomics models for predicting pathological complete response in locally advanced rectal cancer: a two-centre, multi-vendor study

OBJECTIVES

The aim of this study was two-fold: (1) to develop and externally validate a multiparameter MR-based machine learning model to predict the pathological complete response (pCR) in locally advanced rectal cancer (LARC) patients after neoadjuvant chemoradiotherapy (nCRT), and (2) to compare different classifiers' discriminative performance for pCR prediction.

METHODS

This retrospective study includes 151 LARC patients divided into internal (centre A, n = 100) and external validation set (centre B, n = 51). The clinical and MR radiomics features were derived to construct clinical, radiomics, and clinical-radiomics model. Random forest (RF), support vector machine (SVM), logistic regression (LR), K-nearest neighbor (KNN), naive Bayes (NB), and extreme gradient boosting (XGBoost) were used as classifiers. The predictive performance was assessed using the receiver operating characteristic (ROC) curve.

RESULTS

Eleven radiomics and four clinical features were chosen as pCR-related signatures. In the radiomics model, the RF algorithm achieved 74.0% accuracy (an AUC of 0.863) and 84.4% (an AUC of 0.829) in the internal and external validation sets. In the clinical-radiomics model, RF algorithm exhibited high and stable predictive performance in the internal and external validation datasets with an AUC of 0.906 (87.3% sensitivity, 73.7% specificity, 76.0% accuracy) and 0.872 (77.3% sensitivity, 88.2% specificity, 86.3% accuracy), respectively. RF showed a better predictive performance than the other classifiers in the external validation datasets of three models.

CONCLUSIONS

The multiparametric clinical-radiomics model combined with RF algorithm is optimal for predicting pCR in the internal and external sets, and might help improve clinical stratifying management of LARC patients.

KEY POINTS

• A two-centre study showed that radiomics analysis of pre- and post-nCRT multiparameter MR images could predict pCR in patients with LARC. • The combined model was superior to the clinical and radiomics model in predicting pCR in locally advanced rectal cancer. • The RF classifier performed best in the current study.

Metabolism-related lncRNAs signature to predict the prognosis of colon adenocarcinoma

BACKGROUND

Cell metabolism and long noncoding RNA (lncRNA) played crucial roles in cancer development. However, their association in colon adenocarcinoma (COAD) remains unclear.

METHODS

The COAD gene expression data and corresponding clinical data were retrieved from The Cancer Genome Atlas (TCGA) database. Differential expression of metabolic genes and lncRNA were identified by comparing tumor and normal colon tissues. Pearson correlation analysis was performed to identify metabolism-associated lncRNA. COAD patients were divided into training cohort and validation cohort by randomization. Then, a univariate Cox regression analysis was introduced to evaluate the correlations between metabolism-related lncRNAs and overall survival (OS) of the patients in the training cohort. The least absolute shrinkage and selection operator (LASSO) method was introduced to determine and establish a prognostic prediction model. Subsequently, survival analysis, receiver operating characteristic (ROC) curve analysis, and Cox regression analysis were generated to estimate the prognostic role of the LncRNA risk score in training, validation, and entire cohorts.

RESULTS

We identified 152 differentially expressed metabolism-associated lncRNAs (MRLncRNAs). A prognostic prediction model involving four metabolism-related lncRNAs were established using LASSO. In each cohort, COAD patients in the high-risk group had worse OS compared to those in the low-risk group. The ROC analyses demonstrated that the lncRNA signature performed well in predicting OS. Uni- and multivariate analysis indicated that the lncRNA signature as an independent prognostic factor. Furthermore, a correlation analysis demonstrated that LINC01138 was the most closely lncRNA related to metabolic genes. In vitro assays demonstrated that LINC01138 affects tumor progression in COAD.

CONCLUSIONS

In summary, we established a metabolism-associated lncRNAs model to predict the prognosis in COAD patients.

Diagnostic, clustering, and immune cell infiltration analysis of m6A regulators in patients with sepsis

RNA N6-methladenosine (m6A) regulators are required for a variety of biological processes, including immune responses, and increasing evidence indicates that their dysregulation is closely associated with many diseases. However, the potential roles of m6A regulators in sepsis remain unknown. We comprehensively analyzed the transcriptional variations in and interactions of 26 m6A regulators in sepsis based on the Gene Expression Omnibus (GEO) database. A random forest (RF) model and nomogram were established to predict the occurrence and risk of sepsis in patients. Then, two different m6A subtypes were defined by consensus clustering analysis, and we explored the correlation between the subtypes and immune cells. We found that 17 of the 26 m6A regulators were significantly differentially expressed between patients with and without sepsis, and strong correlations among these 17 m6A regulators were revealed. Compared with the support vector machine (SVM) model, the RF model had better predictive ability, and therefore was used to construct a reliable nomogram containing 10 candidate m6A regulators to predict the risk of sepsis in patients. In addition, a consensus clustering algorithm was used to identify two different subtypes of m6A, which helped us distinguish different levels of immune cell infiltration and inflammation in patients with sepsis. Comprehensive analysis of m6A regulators in sepsis revealed their potential roles in sepsis occurrence, immune cell infiltration and inflammation in patients with sepsis. This study may contribute to the development of follow-up treatment strategies for sepsis.

Empirical comparison of routinely collected electronic health record data for head and neck cancer-specific survival in machine-learnt prognostic models

BACKGROUND

Knowledge of the prognostic factors and performance of machine learning predictive models for 2-year cancer-specific survival (CSS) is limited in the head and neck cancer (HNC) population.

METHODS

Data from our facilities' oncology information system (OIS) collected for routine practice (OIS dataset, n = 430 patients) and research purposes (research dataset, n = 529 patients) were extracted on adults diagnosed between 2000 and 2017 with squamous cell carcinoma of the head and neck.

RESULTS

Machine learning demonstrated excellent performance (area under the curve, AUC) in the whole cohort (AUC = 0.97, research dataset), larynx cohort (AUC = 0.98, both datasets), and oropharynx cohort (AUC = 0.99, both datasets). Tumor site and T classification were identified as predictors of 2-year CSS in both datasets. Hypothyroidism and fitness for operation were further identified in the research dataset.

CONCLUSIONS

Datasets extracted from an OIS for routine clinical practice and research purposes demonstrated high utility for informing 2-year head and neck CSS.

Predicting Physiological Response in Heart Failure Management: A Graph Representation Learning Approach using Electronic Health Records

Heart failure management is challenging due to the complex and heterogenous nature of its pathophysiology which makes the conventional treatments based on the "one size fits all" ideology not suitable. Coupling the longitudinal medical data with novel deep learning and network-based analytics will enable identifying the distinct patient phenotypic characteristics to help individualize the treatment regimen through the accurate prediction of the physiological response. In this study, we develop a graph representation learning framework that integrates the heterogeneous clinical events in the electronic health records (EHR) as graph format data, in which the patient-specific patterns and features are naturally infused for personalized predictions of lab test response. The framework includes a novel Graph Transformer Network that is equipped with a self-attention mechanism to model the underlying spatial interdependencies among the clinical events characterizing the cardiac physiological interactions in the heart failure treatment and a graph neural network (GNN) layer to incorporate the explicit temporality of each clinical event, that would help summarize the therapeutic effects induced on the physiological variables, and subsequently on the patient's health status as the heart failure condition progresses over time. We introduce a global attention mask that is computed based on event co-occurrences and is aggregated across all patient records to enhance the guidance of neighbor selection in graph representation learning. We test the feasibility of our model through detailed quantitative and qualitative evaluations on observational EHR data.