Area under the free-response ROC curve (FROC) and a related summary index

Development and external validation of a transfer learning-based system for the pathological diagnosis of colorectal cancer: a large emulated prospective study

Background

The progress in Colorectal cancer (CRC) screening and management has resulted in an unprecedented caseload for histopathological diagnosis. While artificial intelligence (AI) presents a potential solution, the predominant emphasis on slide-level aggregation performance without thorough verification of cancer in each location, impedes both explainability and transparency. Effectively addressing these challenges is crucial to ensuring the reliability and efficacy of AI in histology applications.

Method

In this study, we created an innovative AI algorithm using transfer learning from a polyp segmentation model in endoscopy. The algorithm precisely localized CRC targets within 0.25 mm² grids from whole slide imaging (WSI). We assessed the CRC detection capabilities at this fine granularity and examined the influence of AI on the diagnostic behavior of pathologists. The evaluation utilized an extensive dataset comprising 858 consecutive patient cases with 1418 WSIs obtained from an external center.

Results

Our results underscore a notable sensitivity of 90.25% and specificity of 96.60% at the grid level, accompanied by a commendable area under the curve (AUC) of 0.962. This translates to an impressive 99.39% sensitivity at the slide level, coupled with a negative likelihood ratio of <0.01, signifying the dependability of the AI system to preclude diagnostic considerations. The positive likelihood ratio of 26.54, surpassing 10 at the grid level, underscores the imperative for meticulous scrutiny of any AI-generated highlights. Consequently, all four participating pathologists demonstrated statistically significant diagnostic improvements with AI assistance.

Conclusion

Our transfer learning approach has successfully yielded an algorithm that can be validated for CRC histological localizations in whole slide imaging. The outcome advocates for the integration of the AI system into histopathological diagnosis, serving either as a diagnostic exclusion application or a computer-aided detection (CADe) tool. This integration has the potential to alleviate the workload of pathologists and ultimately benefit patients.

Nomogram for predicting surgical site infections in elderly patients after open lumbar spine surgery: A retrospective study

The aim of this study is to develop a nomogram to assess the risk of surgical site infection in elderly patients undergoing open lumbar spine surgery and explore related risk factors. We reviewed the records of 578 elderly patients who had undergone open lumbar spine surgery. The clinical parameters were subjected to lasso regression and logistic regression analyses. Subsequently, a nomogram was constructed to predict the risk of postoperative surgical site infection and validated using bootstrap resampling. A total of 578 patients were included in the analysis, of which 17 were diagnosed as postoperative surgical site infection. Following the final logistic regression analysis, obesity, hypoalbuminemia and drinking history were identified as independent risk factors and subsequently incorporated into the nomogram. The nomogram demonstrated excellent discrimination, with an area under the receiver-operating characteristic curve of 0.879 (95% CI 0.769 ~ 0.989) after internal validation. The calibration curve exhibited a high level of consistency. Decision curve analysis revealed that this nomogram had greater clinical value when the risk threshold for surgical site infection occurrence was >1% and <89%. We had developed a nomogram for predicting the risk of postoperative surgical site infection in elderly patients who had undergone open lumbar spine surgery. Validation using bootstrap resampling demonstrated excellent discrimination and calibration, indicating that the nomogram may hold potential clinical utility as a simple predictive tool for healthcare professionals.

Understanding metric-related pitfalls in image analysis validation

Validation metrics are key for the reliable tracking of scientific progress and for bridging the current chasm between artificial intelligence (AI) research and its translation into practice. However, increasing evidence shows that particularly in image analysis, metrics are often chosen inadequately in relation to the underlying research problem. This could be attributed to a lack of accessibility of metric-related knowledge: While taking into account the individual strengths, weaknesses, and limitations of validation metrics is a critical prerequisite to making educated choices, the relevant knowledge is currently scattered and poorly accessible to individual researchers. Based on a multi-stage Delphi process conducted by a multidisciplinary expert consortium as well as extensive community feedback, the present work provides the first reliable and comprehensive common point of access to information on pitfalls related to validation metrics in image analysis. Focusing on biomedical image analysis but with the potential of transfer to other fields, the addressed pitfalls generalize across application domains and are categorized according to a newly created, domain-agnostic taxonomy. To facilitate comprehension, illustrations and specific examples accompany each pitfall. As a structured body of information accessible to researchers of all levels of expertise, this work enhances global comprehension of a key topic in image analysis validation.

Performance Analysis in Children of Traditional and Deep Learning CT Lung Nodule Computer-Aided Detection Systems Trained on Adults

BACKGROUND. Although primary lung cancer is rare in children, chest CT is commonly performed to assess for lung metastases in children with cancer. Lung nodule computer-aided detection (CAD) systems have been designed and studied primarily using adult training data, and the efficacy of such systems when applied to pediatric patients is poorly understood. OBJECTIVE. The purpose of this study was to evaluate in children the diagnostic performance of traditional and deep learning CAD systems trained with adult data for the detection of lung nodules on chest CT scans and to compare the ability of such systems to generalize to children versus to other adults. METHODS. This retrospective study included pediatric and adult chest CT test sets. The pediatric test set comprised 59 CT scans in 59 patients (30 boys, 29 girls; mean age, 13.1 years; age range, 4-17 years), which were obtained from November 30, 2018, to August 31, 2020; lung nodules were annotated by fellowship-trained pediatric radiologists as the reference standard. The adult test set was the publicly available adult Lung Nodule Analysis (LUNA) 2016 subset 0, which contained 89 deidentified scans with previously annotated nodules. The test sets were processed through the traditional FlyerScan (github.com/rhardie1/FlyerScanCT) and deep learning Medical Open Network for Artificial Intelligence (MONAI; github.com/Project-MONAI/model-zoo/releases) lung nodule CAD systems, which had been trained on separate sets of CT scans in adults. Sensitivity and false-positive (FP) frequency were calculated for nodules measuring 3-30 mm; nonoverlapping 95% CIs indicated significant differences. RESULTS. Operating at two FPs per scan, on pediatric testing data FlyerScan and MONAI showed significantly lower detection sensitivities of 68.4% (197/288; 95% CI, 65.1-73.0%) and 53.1% (153/288; 95% CI, 46.7-58.4%), respectively, than on adult LUNA 2016 subset 0 testing data (83.9% [94/112; 95% CI, 79.1-88.0%] and 95.5% [107/112; 95% CI, 90.0-98.4%], respectively). Mean nodule size was smaller (p < .001) in the pediatric testing data (5.4 ± 3.1 [SD] mm) than in the adult LUNA 2016 subset 0 testing data (11.0 ± 6.2 mm). CONCLUSION. Adult-trained traditional and deep learning-based lung nodule CAD systems had significantly lower sensitivity for detection on pediatric data than on adult data at a matching FP frequency. The performance difference may relate to the smaller size of pediatric lung nodules. CLINICAL IMPACT. The results indicate a need for pediatric-specific lung nodule CAD systems trained on data specific to pediatric patients.

Construction of a predictive model for blood transfusion in patients undergoing total hip arthroplasty and identification of clinical heterogeneity

A precise forecast of the need for blood transfusions (BT) in patients undergoing total hip arthroplasty (THA) is a crucial step toward the implementation of precision medicine. To achieve this goal, we utilized supervised machine learning (SML) techniques to establish a predictive model for BT requirements in THA patients. Additionally, we employed unsupervised machine learning (UML) approaches to identify clinical heterogeneity among these patients. In this study, we recruited 224 patients undergoing THA. To identify factors predictive of BT during the perioperative period of THA, we employed LASSO regression and the random forest (RF) algorithm as part of supervised machine learning (SML). Using logistic regression, we developed a predictive model for BT in THA patients. Furthermore, we utilized unsupervised machine learning (UML) techniques to cluster THA patients who required BT based on similar clinical features. The resulting clusters were subsequently visualized and validated. We constructed a predictive model for THA patients who required BT based on six predictive factors: Age, Body Mass Index (BMI), Hemoglobin (HGB), Platelet (PLT), Bleeding Volume, and Urine Volume. Before surgery, 1 h after surgery, 1 day after surgery, and 1 week after surgery, significant differences were observed in HGB and PLT levels between patients who received BT and those who did not. The predictive model achieved an AUC of 0.899. Employing UML, we identified two distinct clusters with significantly heterogeneous clinical characteristics. Age, BMI, PLT, HGB, bleeding volume, and urine volume were found to be independent predictors of BT requirement in THA patients. The predictive model incorporating these six predictors demonstrated excellent predictive performance. Furthermore, employing UML enabled us to classify a heterogeneous cohort of THA patients who received BT in a meaningful and interpretable manner.

Improving Tumor-Infiltrating Lymphocytes Score Prediction in Breast Cancer with Self-Supervised Learning

Tumor microenvironment (TME) plays a pivotal role in immuno-oncology, which investigates the intricate interactions between tumors and the human immune system. Specifically, tumor-infiltrating lymphocytes (TILs) are crucial biomarkers for evaluating the prognosis of breast cancer patients and have the potential to refine immunotherapy precision and accurately identify tumor cells in specific cancer types. In this study, we conducted tissue segmentation and lymphocyte detection tasks to predict TIL scores by employing self-supervised learning (SSL) model-based approaches capable of addressing limited labeling data issues. Our experiments showed a 1.9% improvement in tissue segmentation and a 2% improvement in lymphocyte detection over the ImageNet pre-training model. Using these SSL-based models, we achieved a TIL score of 0.718 with a 4.4% improvement. In particular, when trained with only 10% of the entire dataset, the SwAV pre-trained model exhibited a superior performance over other models. Our work highlights improved tissue segmentation and lymphocyte detection using the SSL model with less labeled data for TIL score prediction.

The stage-dependent prognostic role of ARID1A in hepatocellular carcinoma

Background

Hepatocellular carcinoma (HCC) is one of the most common causes of cancer-related death. Although novel treatment currently achieves a better response, the majority of HCC patients develop resistance and cannot benefit. Hence, novel biomarkers for guiding therapy and predicting the prognosis are needed.

Methods

Tissue microarrays of 206 HCC patients were used, and ARID1A expression was determined by immunohistochemistry. Databases were used for the verification and expansion of our results. The "rms" package of R software was used for the construction of the nomogram.

Results

ARID family alterations were associated with disease-free survival (P=0.0325) and overall survival (OS) (P=0.0076). Subgroup analysis confirmed the prognostic effect of ARID1A, ARID1B, and ARID2 alterations. In addition, ARID family genomic alterations, especially ARID1A, were closely related to poor progression-free survival (ARID: P=0.0011; ARID1A: P=0.0082) and OS (ARID: P=0.0161; ARID1A: P=0.0220) after sorafenib treatment. ARID1A expression was found to display a stage-dependent effect on the prognosis, serving as a risk factor in stage I-II patients (P<0.0001) and a protective factor in stage III-IV patients (P=0.0180).

Conclusions

ARID1A has dual roles in HCC in a tumor stage-dependent manner, and further study is required to uncover the complex function of ARID1A in HCC development, disease progression, and therapy.

Construction and validation of a predictive model for postoperative urinary retention after lumbar interbody fusion surgery

BACKGROUND

Postoperative urine retention (POUR) after lumbar interbody fusion surgery may lead to recatheterization and prolonged hospitalization. In this study, a predictive model was constructed and validated. The objective was to provide a nomogram for estimating the risk of POUR and then reducing the incidence.

METHODS

A total of 423 cases of lumbar fusion surgery were included; 65 of these cases developed POUR, an incidence of 15.4%. The dataset is divided into a training set and a validation set according to time. 18 candidate variables were selected. The candidate variables were screened through LASSO regression. The stepwise regression and random forest analysis were then conducted to construct the predictive model and draw a nomogram. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve and the calibration curve were used to evaluate the predictive effect of the model.

RESULTS

The best lambda value in LASSO was 0.025082; according to this, five significant variables were screened, including age, smoking history, surgical method, operative time, and visual analog scale (VAS) score of postoperative low back pain. A predictive model containing four variables was constructed by stepwise regression. The variables included age (β = 0.047, OR = 1.048), smoking history (β = 1.950, OR = 7.031), operative time (β = 0.022, OR = 1.022), and postoperative VAS score of low back pain (β = 2.554, OR = 12.858). A nomogram was drawn based on the results. The AUC of the ROC curve of the training set was 0.891, the validation set was 0.854 in the stepwise regression model. The calibration curves of the training set and validation set are in good agreement with the actual curves, showing that the stepwise regression model has good prediction ability. The AUC of the training set was 0.996, and that of the verification set was 0.856 in the random forest model.

CONCLUSION

This study developed and internally validated a new nomogram and a random forest model for predicting the risk of POUR after lumbar interbody fusion surgery. Both of the nomogram and the random forest model have high accuracy in this study.

Demystifying the impact of prenatal tobacco exposure on the placental immune microenvironment: Avoiding the tragedy of mending the fold after death

Prenatal tobacco exposure (PTE) correlates significantly with a surge in adverse pregnancy outcomes, yet its pathological mechanisms remain partially unexplored. This study aims to meticulously examine the repercussions of PTE on placental immune landscapes, employing a coordinated research methodology encompassing bioinformatics, machine learning and animal studies. Concurrently, it aims to screen biomarkers and potential compounds that could sensitively indicate and mitigate placental immune disorders. In the course of this research, two gene expression omnibus (GEO) microarrays, namely GSE27272 and GSE7434, were included. Gene set enrichment analysis (GSEA) and immune enrichment investigations on differentially expressed genes (DEGs) indicated that PTE might perturb numerous innate or adaptive immune-related biological processes. A cohort of 52 immune-associated DEGs was acquired by cross-referencing the DEGs with gene sets derived from the ImmPort database. A protein-protein interaction (PPI) network was subsequently established, from which 10 hub genes were extracted using the maximal clique centrality (MCC) algorithm (JUN, NPY, SST, FLT4, FGF13, HBEGF, NR0B2, AREG, NR1I2, SEMA5B). Moreover, we substantiated the elevated affinity of tobacco reproductive toxicants, specifically nicotine and nitrosamine, with hub genes through molecular docking (JUN, FGF13 and NR1I2). This suggested that these genes could potentially serve as crucial loci for tobacco's influence on the placental immune microenvironment. To further elucidate the immune microenvironment landscape, consistent clustering analysis was conducted, yielding three subtypes, where the abundance of follicular helper T cells (p < 0.05) in subtype A, M2 macrophages (p < 0.01), neutrophils (p < 0.05) in subtype B and CD8+ T cells (p < 0.05), resting NK cells (p < 0.05), M2 macrophages (p < 0.05) in subtype C were significantly different from the control group. Additionally, three pivotal modules, designated as red, blue and green, were identified, each bearing a close association with differentially infiltrated immunocytes, as discerned by the weighted gene co-expression network analysis (WGCNA). Functional enrichment analysis was subsequently conducted on these modules. To further probe into the mechanisms by which immune-associated DEGs are implicated in intercellular communication, 20 genes serving as ligands or receptors and connected to differentially infiltrating immunocytes were isolated. Employing a variety of machine learning techniques, including one-way logistic regression, LASSO regression, random forest and artificial neural networks, we screened 11 signature genes from the intersection of immune-associated DEGs and secretory protein-encoding genes derived from the Human Protein Atlas. Notably, CCL18 and IFNA4 emerged as prospective peripheral blood markers capable of identifying PTE-induced immune disorders. These markers demonstrated impressive predictive power, as indicated by the area under the curve (AUC) of 0.713 (0.548-0.857) and 0.780 (0.618-0.914), respectively. Furthermore, we predicted 34 potential compounds, including cyclosporine, oestrogen and so on, which may engage with hub genes and attenuate immune disorders instigated by PTE. The diagnostic performance of these biomarkers, alongside the interventional effect of cyclosporine, was further corroborated in animal studies via ELISA, Western blot and immunofluorescence assays. In summary, this study identifies a disturbance in the placental immune landscape, a secondary effect of PTE, which may underlie multiple pregnancy complications. Importantly, our research contributes to the noninvasive and timely detection of PTE-induced placental immune disorders, while also offering innovative therapeutic strategies for their treatment.

Weakly supervised breast lesion detection in DCE-MRI using self-transfer learning

BACKGROUND

Breast cancer is a typically diagnosed and life-threatening cancer in women. Thus, dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is increasingly used for breast lesion detection and diagnosis because of the high resolution of soft tissues. Moreover, supervised detection methods have been implemented for breast lesion detection. However, these methods require substantial time and specialized staff to develop the labeled training samples.

PURPOSE

To investigate the potential of weakly supervised deep learning models for breast lesion detection.

METHODS

A total of 1003 breast DCE-MRI studies were collected, including 603 abnormal cases with 770 breast lesions and 400 normal subjects. The proposed model was trained using breast DCE-MRI considering only the image-level labels (normal and abnormal) and optimized for classification and detection sub-tasks simultaneously. Ablation experiments were performed to evaluate different convolutional neural network (CNN) backbones (VGG19 and ResNet50) as shared convolutional layers, as well as to evaluate the effect of the preprocessing methods.

RESULTS

Our weakly supervised model performed better with VGG19 than with ResNet50 (p < 0.05). The average precision (AP) of the classification sub-task was 91.7% for abnormal cases and 88.0% for normal samples. The area under the receiver operating characteristic (ROC) curve (AUC) was 0.939 (95% confidence interval [CI]: 0.920-0.941). The weakly supervised detection task AP was 85.7%, and the correct location (CorLoc) was 90.2%. A sensitivity of 84.0% at two-false positives per image was assessed based on free-response ROC (FROC) curve.

CONCLUSIONS

The results confirm that a weakly supervised CNN based on self-transfer learning is an effective and promising auxiliary tool for detecting breast lesions.

Computer-aided diagnosis systems: a comparative study of classical machine learning versus deep learning-based approaches

The diagnostic phase of the treatment process is essential for patient guidance and follow-up. The accuracy and effectiveness of this phase can determine the life or death of a patient. For the same symptoms, different doctors may come up with different diagnoses whose treatments may, instead of curing a patient, be fatal. Machine learning (ML) brings new solutions to healthcare professionals to save time and optimize the appropriate diagnosis. ML is a data analysis method that automates the creation of analytical models and promotes predictive data. There are several ML models and algorithms that rely on features extracted from, for example, a patient's medical images to indicate whether a tumor is benign or malignant. The models differ in the way they operate and the method used to extract the discriminative features of the tumor. In this article, we review different ML models for tumor classification and COVID-19 infection to evaluate the different works. The computer-aided diagnosis (CAD) systems, which we referred to as classical, are based on accurate feature identification, usually performed manually or with other ML techniques that are not involved in classification. The deep learning-based CAD systems automatically perform the identification and extraction of discriminative features. The results show that the two types of DAC have quite close performances but the use of one or the other type depends on the datasets. Indeed, manual feature extraction is necessary when the size of the dataset is small; otherwise, deep learning is used.

A Novel Deep Learning Model Based on Multi-Scale and Multi-View for Detection of Pulmonary Nodules

Lung cancer manifests as pulmonary nodules in the early stage. Thus, the early and accurate detection of these nodules is crucial for improving the survival rate of patients. We propose a novel two-stage model for lung nodule detection. In the candidate nodule detection stage, a deep learning model based on 3D context information roughly segments the nodules detects the preprocessed image and obtain candidate nodules. In this model, 3D image blocks are input into the constructed model, and it learns the contextual information between the various slices in the 3D image block. The parameters of our model are equivalent to those of a 2D convolutional neural network (CNN), but the model could effectively learn the 3D context information of the nodules. In the false-positive reduction stage, we propose a multi-scale shared convolutional structure model. Our lung detection model has no significant increase in parameters and computation in both stages of multi-scale and multi-view detection. The proposed model was evaluated by using 888 computed tomography (CT) scans from the LIDC-IDRI dataset and achieved a competition performance metric (CPM) score of 0.957. The average detection sensitivity per scan was 0.971/1.0 FP. Furthermore, an average detection sensitivity of 0.933/1.0 FP per scan was achieved based on data from Shanghai Pulmonary Hospital. Our model exhibited a higher detection sensitivity, a lower false-positive rate, and better generalization than current lung nodule detection methods. The method has fewer parameters and less computational complexity, which provides more possibilities for the clinical application of this method.

Mass detection in automated 3-D breast ultrasound using a patch Bi-ConvLSTM network

Breast cancer mortality can be significantly reduced by early detection of its symptoms. The 3-D Automated Breast Ultrasound (ABUS) has been widely used for breast screening due to its high sensitivity and reproducibility. The large number of ABUS slices, and high variation in size and shape of the masses, make the manual evaluation a challenging and time-consuming process. To assist the radiologists, we propose a convolutional BiLSTM network to classify the slices based on the presence of a mass. Because of its patch-based architecture, this model produces the approximate location of masses as a heat map. The prepared dataset consists of 60 volumes belonging to 43 patients. The precision, recall, accuracy, F1-score, and AUC of the proposed model for slice classification were 84%, 84%, 93%, 84%, and 97%, respectively. Based on the FROC analysis, the proposed detector obtained a sensitivity of 82% with two false positives per volume.

Immunogenic Cell Death Associated Molecular Patterns and the Dual Role of IL17RA in Interstitial Cystitis/Bladder Pain Syndrome

The unclear etiology and pathogenesis of interstitial cystitis/bladder pain syndrome (IC/BPS) are responsible for the lack of effective treatment and the poor patient prognosis. Various studies show that chronic inflammation and immune responses are important factors contributing to the pathogenesis of IC/BPS. The process of immunogenic cell death (ICD) involves both the immune response and inflammatory process, and the involvement of ICD in IC/BPS pathogenesis has not been explored. Two IC/BPS transcriptome datasets collected from the Gene Expression Omnibus (GEO) database were used to identify distinct ICD-associated molecular patterns (IAMPs). IAMPs and IC/BPS subtypes were found to be related. The inflammatory immune microenvironments (IIME) in different IAMPs were studied. The potential mechanism by which the interleukin 17 receptor A (IL17RA) influences IC/BPS was examined using in vitro assays. The expression of ICD-related genes (IRGs) was upregulated in IC/BPS bladders, compared with normal bladders. Disease prediction models, based on differentially expressed IRGs, could accurately predict IC/BPS. The IC/BPS patients had two distinct IAMPs, each with its own subtype and clinical features and association with remodeling IIME. IL17RA, a well-established IC/BPS bladder biomarker, mediates both the inflammatory insult and the protective responses. In summary, the current study identified different IAMPs in IC/BPS, which may be involved in the pathogenesis of IC/BPS by remodeling the IIME. The chronic inflammatory process in IC/BPS may be prolonged by IL17RA, which could mediate both pro- and anti-inflammatory responses. The IL17RA-associated pathway may play a significant role in the development of IC/BPS and can be used as a therapeutic target.

The Low Rate of Adherence to Checklist for Artificial Intelligence in Medical Imaging Criteria Among Published Prostate MRI Artificial Intelligence Algorithms

OBJECTIVE

To determine the rigor, generalizability, and reproducibility of published classification and detection artificial intelligence (AI) models for prostate cancer (PCa) on MRI using the Checklist for Artificial Intelligence in Medical Imaging (CLAIM) guidelines, a 42-item checklist that is considered a measure of best practice for presenting and reviewing medical imaging AI research.

MATERIALS AND METHODS

This review searched English literature for studies proposing PCa AI detection and classification models on MRI. Each study was evaluated with the CLAIM checklist. The additional outcomes for which data were sought included measures of AI model performance (eg, area under the curve [AUC], sensitivity, specificity, free-response operating characteristic curves), training and validation and testing group sample size, AI approach, detection versus classification AI, public data set utilization, MRI sequences used, and definition of gold standard for ground truth. The percentage of CLAIM checklist fulfillment was used to stratify studies into quartiles. Wilcoxon's rank-sum test was used for pair-wise comparisons.

RESULTS

In all, 75 studies were identified, and 53 studies qualified for analysis. The original CLAIM items that most studies did not fulfill includes item 12 (77% no): de-identification methods; item 13 (68% no): handling missing data; item 15 (47% no): rationale for choosing ground truth reference standard; item 18 (55% no): measurements of inter- and intrareader variability; item 31 (60% no): inclusion of validated interpretability maps; item 37 (92% no): inclusion of failure analysis to elucidate AI model weaknesses. An AUC score versus percentage CLAIM fulfillment quartile revealed a significant difference of the mean AUC scores between quartile 1 versus quartile 2 (0.78 versus 0.86, P = .034) and quartile 1 versus quartile 4 (0.78 versus 0.89, P = .003) scores. Based on additional information and outcome metrics gathered in this study, additional measures of best practice are defined. These new items include disclosure of public dataset usage, ground truth definition in comparison to other referenced works in the defined task, and sample size power calculation.

CONCLUSION

A large proportion of AI studies do not fulfill key items in CLAIM guidelines within their methods and results sections. The percentage of CLAIM checklist fulfillment is weakly associated with improved AI model performance. Additions or supplementations to CLAIM are recommended to improve publishing standards and aid reviewers in determining study rigor.

High-resolution synthesis of high-density breast mammograms: Application to improved fairness in deep learning based mass detection

Computer-aided detection systems based on deep learning have shown good performance in breast cancer detection. However, high-density breasts show poorer detection performance since dense tissues can mask or even simulate masses. Therefore, the sensitivity of mammography for breast cancer detection can be reduced by more than 20% in dense breasts. Additionally, extremely dense cases reported an increased risk of cancer compared to low-density breasts. This study aims to improve the mass detection performance in high-density breasts using synthetic high-density full-field digital mammograms (FFDM) as data augmentation during breast mass detection model training. To this end, a total of five cycle-consistent GAN (CycleGAN) models using three FFDM datasets were trained for low-to-high-density image translation in high-resolution mammograms. The training images were split by breast density BI-RADS categories, being BI-RADS A almost entirely fatty and BI-RADS D extremely dense breasts. Our results showed that the proposed data augmentation technique improved the sensitivity and precision of mass detection in models trained with small datasets and improved the domain generalization of the models trained with large databases. In addition, the clinical realism of the synthetic images was evaluated in a reader study involving two expert radiologists and one surgical oncologist.

Automated meniscus segmentation and tear detection of knee MRI with a 3D mask-RCNN

BACKGROUND

The diagnostic results of magnetic resonance imaging (MRI) are essential references for arthroscopy as an invasive procedure. A deviation between medical imaging diagnosis and arthroscopy results may cause irreversible damage to patients and lead to excessive medical treatment. To improve the accurate diagnosis of meniscus injury, it is urgent to develop auxiliary diagnosis algorithms to improve the accuracy of radiological diagnosis.

PURPOSE

This study aims to present a fully automatic 3D deep convolutional neural network (DCNN) for meniscus segmentation and detects arthroscopically proven meniscus tears.

MATERIALS AND METHODS

Our institution retrospectively included 533 patients with 546 knees who underwent knee magnetic resonance imaging (MRI) and knee arthroscopy. Sagittal proton density-weighted (PDW) images in MRI of 382 knees were regarded as a training set to train our 3D-Mask RCNN. The remaining data from 164 knees were used to validate the trained network as a test set. The masks were hand-drawn by an experienced radiologist, and the reference standard is arthroscopic surgical reports. The performance statistics included Dice accuracy, sensitivity, specificity, FROC, receiver operating characteristic (ROC) curve analysis, and bootstrap test statistics. The segmentation performance was compared with a 3D-Unet, and the detection performance was compared with radiological evaluation by two experienced musculoskeletal radiologists without knowledge of the arthroscopic surgical diagnosis.

RESULTS

Our model produced strong Dice coefficients for sagittal PDW of 0.924, 0.95 sensitivity with 0.823 FPs/knee. 3D-Unet produced a Dice coefficient for sagittal PDW of 0.891, 0.95 sensitivity with 1.355 FPs/knee. The difference in the areas under 3D-Mask-RCNN FROC and 3D-Unet FROC was statistically significant (p = 0.0011) by bootstrap test. Our model detection performance achieved an area under the curve (AUC) value, accuracy, and sensitivity of 0.907, 0.924, 0.941, and 0.785, respectively. Based on the radiological evaluations, the AUC value, accuracy, sensitivity, and specificity were 0.834, 0.835, 0.889, and 0.754, respectively. The difference in the areas between 3D-Mask-RCNN ROC and radiological evaluation ROC was statistically significant (p = 0.0009) by bootstrap test. 3D Mask RCNN significantly outperformed the 3D-Unet and radiological evaluation demonstrated by these results.

CONCLUSIONS

3D-Mask RCNN has demonstrated efficacy and precision for meniscus segmentation and tear detection in knee MRI, which can assist radiologists in improving the accuracy and efficiency of diagnosis. It can also provide effective diagnostic indicators for orthopedic surgeons before arthroscopic surgery and further promote precise treatment.

Domain generalization in deep learning based mass detection in mammography: A large-scale multi-center study

Computer-aided detection systems based on deep learning have shown great potential in breast cancer detection. However, the lack of domain generalization of artificial neural networks is an important obstacle to their deployment in changing clinical environments. In this study, we explored the domain generalization of deep learning methods for mass detection in digital mammography and analyzed in-depth the sources of domain shift in a large-scale multi-center setting. To this end, we compared the performance of eight state-of-the-art detection methods, including Transformer based models, trained in a single domain and tested in five unseen domains. Moreover, a single-source mass detection training pipeline was designed to improve the domain generalization without requiring images from the new domain. The results show that our workflow generalized better than state-of-the-art transfer learning based approaches in four out of five domains while reducing the domain shift caused by the different acquisition protocols and scanner manufacturers. Subsequently, an extensive analysis was performed to identify the covariate shifts with the greatest effects on detection performance, such as those due to differences in patient age, breast density, mass size, and mass malignancy. Ultimately, this comprehensive study provides key insights and best practices for future research on domain generalization in deep learning based breast cancer detection.

Cuproptosis Combined with lncRNAs Predicts the Prognosis and Immune Microenvironment of Breast Cancer

Breast cancer (BC), the most common cancer in women, is caused by the uncontrolled proliferation of mammary epithelial cells under the action of a variety of carcinogenic factors. Cuproptosis-related targets have been found to be closely associated with breast cancer development. TCGA obtained 1226 tumor samples, 1073 clinical data, and 37 lncRNAs during univariate Cox multivariate analysis. We used nonnegative matrix factoring (NMF) agglomeration to spot thirty-three potential molecular subsets with totally different cuproptosis-related lncRNA expression patterns. The least absolute shrinkage and selection operator (LASSO) formula and variable Cox multivariate analysis were not used to construct the best prognostic model. The variations in neoplasm mutation burden and factor gene ontology (GO) and gene set enrichment analysis (GSEA) within the high- and low-risk teams were analyzed, and therefore, the potential mechanism of the development of carcinoma was analyzed. We created a prognostic profile consisting of nineteen cuproptosis-related genes (NFE2L2, LIPT1, LIPT2, DLD, etc.) and their connected targets. The correlation between tumor mutational burden (TMB) and clinical manifestations of tumors demonstrates the importance of high- and low-expression bunch data on the incidence of clinical manifestations of tumors. The area under the curve (AUC) shows moderate prophetic power for copper mortality. GO enrichment analysis showed that immunorelated responses were enriched. Correlation analysis of immune cells showed that pathology could play an important role in the prevalence and prognosis of tumors, and there were variations in immune cells between the probable and low-risk groups. Our study suggests that the prognostic characteristic genes associated with cuproptosis can be used as new biomarkers to predict the prognosis of breast cancer patients. In addition, we found that immunotherapy may play a key role in breast cancer treatment regimens. Levels of immune-associated cells and pathways vary significantly among risk groups of breast cancer patients.

A ubiquitin-related gene signature for predicting prognosis and constructing molecular subtypes in osteosarcoma

Background: Ubiquitination is medicated by three classes of enzymes and has been proven to involve in multiple cancer biological processes. Moreover, dysregulation of ubiquitination has received a growing body of attention in osteosarcoma (OS) tumorigenesis and treatment. Therefore, our study aimed to identify a ubiquitin-related gene signature for predicting prognosis and immune landscape and constructing OS molecular subtypes.

Methods: Therapeutically Applicable Research to Generate Effective Treatments (TARGET) was regarded as the training set through univariate Cox regression, Lasso Cox regression, and multivariate Cox regression. The GSE21257 and GSE39055 served as the validation set to verify the predictive value of the signature. CIBERSORT was performed to show immune infiltration and the immune microenvironment. The NMF algorithm was used to construct OS molecular subtypes.

Results: In this study, we developed a ubiquitin-related gene signature including seven genes (UBE2L3, CORO6, DCAF8, DNAI1, FBXL5, UHRF2, and WDR53), and the gene signature had a good performance in predicting prognosis for OS patients (AUC values at 1/3/5 years were 0.957, 0.890, and 0.919). Multivariate Cox regression indicated that the risk score model and prognosis stage were also independent prognostic prediction factors. Moreover, analyses of immune cells and immune-related functions showed a significant difference in different risk score groups and the three clusters. The drug sensitivity suggested that IC50 of proteasome inhibitor (MG-132) showed a notable significance between the risk score groups (p < 0.05). Through the NMF algorithm, we obtained the three clusters, and cluster 3 showed better survival outcomes. The expression of ubiquitin-related genes (CORO6, UBE2L3, FBXL5, DNAI1, and DCAF8) showed an obvious significance in normal and osteosarcoma tissues.

Conclusion: We developed a novel ubiquitin-related gene signature which showed better predictive prognostic ability for OS and provided additional information on chemotherapy and immunotherapy. The OS molecular subtypes would also give a useful guide for individualized therapy.

Sampling Based Tumor Recognition in Whole-Slide Histology Image With Deep Learning Approaches

Histopathological identification of tumor tissue is one of the routine pathological diagnoses for pathologists. Recently, computational pathology has been successfully interpreted by a variety of deep learning-based applications. Nevertheless, the high-efficient and spatial-correlated processing of individual patches have always attracted attention in whole-slide image (WSI) analysis. In this paper, we propose a high-throughput system to detect tumor regions in colorectal cancer histology slides precisely. We train a deep convolutional neural network (CNN) model and design a Monte Carlo (MC) adaptive sampling method to estimate the most representative patches in a WSI. Two conditional random field (CRF) models are designed, namely the correction CRF and the prediction CRF are integrated for spatial dependencies of patches. We use three datasets of colorectal cancer from The Cancer Genome Atlas (TCGA) to evaluate the performance of the system. The overall diagnostic time can be reduced from 56.7 percent to 71.7 percent on the slides of a varying tumor distribution, with an increase in classification accuracy.

A novel regression method for the analysis of multireader multicase-free-response receiver operating characteristics studies

In diagnostic radiology, the multireader multicase (MRMC) design and the free-response receiver operating characteristics (FROC) method are often used in combination. The cross-correlated data generated by the MRMC-FROC study leads to difficulties in the corresponding analysis, and the need to include covariates in the model further complicates the subsequent analysis. In this paper, we propose a regression approach based on three new measures with good interpretability. The correlation structure of the original test results is taken directly into account in the estimation procedure. The proposed method also allows the inclusion of continuous or discrete covariates. Consistent and asymptotically normal estimators are derived for the new measures. Simulation studies are conducted to evaluate the performance of the proposed approach. The simulation results show that the regression approach performs well under a wide range of scenarios. We also apply the proposed regression approach to a diagnostic study of computer-aided diagnosis in lung cancer.

Identification of an Iron Metabolism-Related lncRNA Signature for Predicting Osteosarcoma Survival and Immune Landscape

Background: Long noncoding RNAs (lncRNAs) act as epigenetic regulators in the process of ferroptosis and iron metabolism. This study aimed to identify an iron metabolism-related lncRNA signature to predict osteosarcoma (OS) survival and the immune landscape.

Methods: RNA-sequencing data and clinical information were obtained from the TARGET dataset. Univariate Cox regression and LASSO Cox analysis were used to develop an iron metabolism-related lncRNA signature. Consensus clustering analysis was applied to identify subtype-based prognosis-related lncRNAs. CIBERSORT was used to analyze the difference in immune infiltration and the immune microenvironment in the two clusters.

Results: We identified 302 iron metabolism-related lncRNAs based on 515 iron metabolism-related genes. The results of consensus clustering showed the differences in immune infiltration and the immune microenvironment in the two clusters. Through univariate Cox regression and LASSO Cox regression analysis, we constructed an iron metabolism-related lncRNA signature that included seven iron metabolism-related lncRNAs. The signature was verified to have good performance in predicting the overall survival, immune-related functions, and immunotherapy response of OS patients between the high- and low-risk groups.

Conclusion: We identified an iron metabolism-related lncRNA signature that had good performance in predicting survival outcomes and showing the immune landscape for OS patients. Furthermore, our study will provide valuable information to further develop immunotherapies of OS.

Multireader Diagnostic Accuracy Imaging Studies: Fundamentals of Design and Analysis

The design and analysis of multireader multicase (MRMC) studies are quite challenging. These studies differ from most medical studies because they need a reference standard and sampling from two populations (ie, reader and patient populations). They are quite expensive to conduct, requiring a good deal of readers' time for image interpretation. One common problem is the use of imperfect reference standards, often correlated with the test or tests being evaluated. Another common issue is oversimplification of the multidimensional MRMC data. In this study, the fundamentals of MRMC study design and analysis are reviewed. The goal is to provide investigators with a guide to the fundamentals of MRMC design and analysis, with references to more detailed discussions. In addition, readers are updated on newer areas of research, including correction for studies with multiple diagnostic accuracy end points and adjustment for location bias.

Development and Validation of a Modified Three-Dimensional U-Net Deep-Learning Model for Automated Detection of Lung Nodules on Chest CT Images From the Lung Image Database Consortium and Japanese Datasets

RATIONALE AND OBJECTIVES

A more accurate lung nodule detection algorithm is needed. We developed a modified three-dimensional (3D) U-net deep-learning model for the automated detection of lung nodules on chest CT images. The purpose of this study was to evaluate the accuracy of the developed modified 3D U-net deep-learning model.

MATERIALS AND METHODS

In this Health Insurance Portability and Accountability Act-compliant, Institutional Review Board-approved retrospective study, the 3D U-net based deep-learning model was trained using the Lung Image Database Consortium and Image Database Resource Initiative dataset. For internal model validation, we used 89 chest CT scans that were not used for model training. For external model validation, we used 450 chest CT scans taken at an urban university hospital in Japan. Each case included at least one nodule of >5 mm identified by an experienced radiologist. We evaluated model accuracy using the competition performance metric (CPM) (average sensitivity at 1/8, 1/4, 1/2, 1, 2, 4, and 8 false-positives per scan). The 95% confidence interval (CI) was computed by bootstrapping 1000 times.

RESULTS

In the internal validation, the CPM was 94.7% (95% CI: 89.1%-98.6%). In the external validation, the CPM was 83.3% (95% CI: 79.4%-86.1%).

CONCLUSION

The modified 3D U-net deep-learning model showed high performance in both internal and external validation.

Development and External Validation of a Nomogram and a Risk Table for Prediction of Type 2 Diabetic Kidney Disease Progression Based on a Retrospective Cohort Study in China

PURPOSE

Diabetic kidney disease (DKD) is the leading cause of end-stage renal disease worldwide. Risk assessment provides information about patient prognosis, contributing to the risk stratification of patients and the rational allocation of medical resources. We aimed to develop a model for individualized prediction of renal function decline in patients with type 2 DKD (T2DKD).

PATIENTS AND METHODS

In a retrospective observational study, we followed 307 T2DKD patients and evaluated the determinants of 1) risk of doubling in serum creatinine (Scr), 2) risk of eGFR<15 mL/min/1.73m² using potential risk factors at baseline. A prediction model represented by a nomogram and a risk table was developed using Cox regression and externally validated in another cohort with 206 T2DKD patients. The discrimination and calibration of the prediction model were evaluated by the concordance index (C-index) and calibration curve, respectively.

RESULTS

Four predictors were selected to establish the final model: Scr, urinary albumin/creatinine ratio, plasma albumin, and insulin treatment. The nomogram achieved satisfactory prediction performance, with a C-index of 0.791 [95% confidence interval (CI) 0.762-0.820] in the derivation cohort and 0.793 (95% CI 0.746-0.840) in the external validation cohort. Then, all predictors were scored according to their weightings. A risk table with the highest score of 11.5 was developed. The C-index of the risk table was 0.764 (95% CI: 0.731-0.797), which was similar to the external validation cohort (0.763; 95% CI: 0.714-0.812). Additionally, the patients were divided into two groups based on the risk table, and significant differences in the probability of outcome events were observed between the high-risk (score >2) and low-risk (score ≤2) groups in the derivation and external validation cohorts (P < 0.001).

CONCLUSION

The nomogram and the risk table using readily available clinical parameters could be new tools for bedside prediction of renal function decline in T2DKD patients.

Performance of a deep learning-based lung nodule detection system as an alternative reader in a Chinese lung cancer screening program

OBJECTIVE

To evaluate the performance of a deep learning-based computer-aided detection (DL-CAD) system in a Chinese low-dose CT (LDCT) lung cancer screening program.

MATERIALS AND METHODS

One-hundred-and-eighty individuals with a lung nodule on their baseline LDCT lung cancer screening scan were randomly mixed with screenees without nodules in a 1:1 ratio (total: 360 individuals). All scans were assessed by double reading and subsequently processed by an academic DL-CAD system. The findings of double reading and the DL-CAD system were then evaluated by two senior radiologists to derive the reference standard. The detection performance was evaluated by the Free Response Operating Characteristic curve, sensitivity and false-positive (FP) rate. The senior radiologists categorized nodules according to nodule diameter, type (solid, part-solid, non-solid) and Lung-RADS.

RESULTS

The reference standard consisted of 262 nodules ≥ 4 mm in 196 individuals; 359 findings were considered false positives. The DL-CAD system achieved a sensitivity of 90.1% with 1.0 FP/scan for detection of lung nodules regardless of size or type, whereas double reading had a sensitivity of 76.0% with 0.04 FP/scan (P = 0.001). The sensitivity for detection of nodules ≥ 4 - ≤ 6 mm was significantly higher with DL-CAD than with double reading (86.3% vs. 58.9% respectively; P = 0.001). Sixty-three nodules were only identified by the DL-CAD system, and 27 nodules only found by double reading. The DL-CAD system reached similar performance compared to double reading in Lung-RADS 3 (94.3% vs. 90.0%, P = 0.549) and Lung-RADS 4 nodules (100.0% vs. 97.0%, P = 1.000), but showed a higher sensitivity in Lung-RADS 2 (86.2% vs. 65.4%, P < 0.001).

CONCLUSIONS

The DL-CAD system can accurately detect pulmonary nodules on LDCT, with an acceptable false-positive rate of 1 nodule per scan and has higher detection performance than double reading. This DL-CAD system may assist radiologists in nodule detection in LDCT lung cancer screening.

Rib fracture detection system based on deep learning

Rib fracture detection is time-consuming and demanding work for radiologists. This study aimed to introduce a novel rib fracture detection system based on deep learning which can help radiologists to diagnose rib fractures in chest computer tomography (CT) images conveniently and accurately. A total of 1707 patients were included in this study from a single center. We developed a novel rib fracture detection system on chest CT using a three-step algorithm. According to the examination time, 1507, 100 and 100 patients were allocated to the training set, the validation set and the testing set, respectively. Free Response ROC analysis was performed to evaluate the sensitivity and false positivity of the deep learning algorithm. Precision, recall, F1-score, negative predictive value (NPV) and detection and diagnosis were selected as evaluation metrics to compare the diagnostic efficiency of this system with radiologists. The radiologist-only study was used as a benchmark and the radiologist-model collaboration study was evaluated to assess the model's clinical applicability. A total of 50,170,399 blocks (fracture blocks, 91,574; normal blocks, 50,078,825) were labelled for training. The F1-score of the Rib Fracture Detection System was 0.890 and the precision, recall and NPV values were 0.869, 0.913 and 0.969, respectively. By interacting with this detection system, the F1-score of the junior and the experienced radiologists had improved from 0.796 to 0.925 and 0.889 to 0.970, respectively; the recall scores had increased from 0.693 to 0.920 and 0.853 to 0.972, respectively. On average, the diagnosis time of radiologist assisted with this detection system was reduced by 65.3 s. The constructed Rib Fracture Detection System has a comparable performance with the experienced radiologist and is readily available to automatically detect rib fracture in the clinical setting with high efficacy, which could reduce diagnosis time and radiologists' workload in the clinical practice.

Loss of ARID1A expression promotes lung adenocarcinoma metastasis and predicts a poor prognosis

BACKGROUND

ARID1A is an essential subunit of SWI/SNF chromatin remodeling complexes. ARID1A gene mutations and loss of ARID1A expression have been observed in a variety of cancers, and to be correlated with invasion, immune escape and synthetic lethality. As yet, however, the biological effect of ARID1A expression and its role in the prognosis of lung adenocarcinoma (LUAD) patients have remained unclear. In this study we aimed to further elucidate the role of ARID1A expression in LUAD in vitro and in vivo and to assess its effect on the clinical prognosis of LUAD patients.

METHODS

ARID1A expression was detected by IHC in tissue samples from LUAD patients. After regular culturing of LUAD cell lines and constructing stable ARID1A knockdown lines, wound healing and Transwell assays were used to assess the role of ARID1A in cell migration and invasion. The effect of ARID1A knockdown on metastasis was verified in vivo. Western blotting was used to examine the expression of target proteins. Univariate and multivariate analyses were performed to assess survival and to provide variables for nomogram construction. In addition, we used the "rms" package to construct a prognostic nomogram based on a Cox regression model.

RESULTS

We found that ARID1A expression serves as an effective prognostic marker for LUAD patients. Loss of ARID1A expression correlated with a poor prognosis, as verified with a nomogram based on a Cox regression model. In addition, we found that ARID1A knockdown promoted LUAD cell proliferation, migration and invasion in vitro and enhanced LUAD metastasis in vivo by activating the Akt signaling pathway.

CONCLUSIONS

Our data indicate that loss of ARID1A expression promotes LUAD metastasis and predicts a poor prognosis in LUAD patients.

A situational awareness Bayesian network approach for accurate and credible personalized adaptive radiotherapy outcomes prediction in lung cancer patients

PURPOSE

A situational awareness Bayesian network (SA-BN) approach is developed to improve physicians' trust in the prediction of radiation outcomes and evaluate its performance for personalized adaptive radiotherapy (pART).

METHODS

118 non-small-cell lung cancer patients with their biophysical features were employed for discovery (n = 68) and validation (n = 50) of radiation outcomes prediction modeling. Patients' important characteristics identified by radiation experts to predict individual's tumor local control (LC) or radiation pneumonitis with grade ≥ 2 (RP2) were incorporated as expert knowledge (EK). Besides generating an EK-based naïve BN (EK-NBN), an SA-BN was developed by incorporating the EK features into pure data-driven BN (PD-BN) methods to improve the credibility of LC or / and RP2 prediction. After using area under the free-response receiver operating characteristics curve (AU-FROC) to assess the joint prediction of these outcomes, their prediction performances were compared with a regression approach based on the expert yielded estimates (EYE) penalty and its variants.

RESULTS

In addition to improving the credibility of radiation outcomes prediction, the SA-BN approach outperformed the EYE penalty and its variants in terms of the joint prediction of LC and RP2. The value of AU-FROC improves from 0.70 (95% CI: 0.54-0.76) using EK-NBN, to 0.75 (0.65-0.82) using a variant of EYE penalty, to 0.83 (0.75-0.93) using PD-BN and 0.83 (0.77-0.90) using SA-BN; with similar trends in the validation cohort.

CONCLUSIONS

The SA-BN approach can provide an accurate and credible human-machine interface to gain physicians' trust in clinical decision-making, which has the potential to be an important component of pART.

A prediction model of simple echocardiographic variables to screen for potentially correctable shunts in adult patients with pulmonary arterial hypertension associated with atrial septal defects: a cross-sectional study

During the routine follow-up of adult patients with pulmonary arterial hypertension associated with atrial septal defects (ASD-PAH), the suitability of shunt closure depends on the invasive right heart catheterization (RHC). It is difficult to grasp the timing of RHC shunt closure for moderate-severe PAH. This retrospective cross-sectional study was designed to investigate which echocardiographic variables are related to pulmonary vascular resistance (PVR) in adult ASD-PAH patients and propose a method using echocardiographic variables to screen for patients where shunt closure is suitable. A total of 139 adult ASD-PAH patients with a PASP ≥ 60 mmHg measured by transthoracic echocardiogram (TTE) were included in this study. All RHCs were performed within a week after TTE. The Correctable shunt was defined as PVR ≤ 4.6 wood units (WU). Multivariate regressions were performed with echocardiographic variables. The nomogram of prediction model was constructed by the predictors of PVR ≤ 4.6 WU by multivariate logistic regression analysis. Multivariate linear regression revealed that TAPSE (tricuspid annular plane systolic excursion)/pulmonary artery systolic pressure (PASP) measured by TTE was negatively associated with PVR (β per SD: - 1.84, 95%CI - 2.62, - 1.06). Multivariate logistic regression showed that TAPSE/PASP and pulmonary valve (PV) peak velocity were positively associated with a potentially correctable shunt (PVR ≤ 4.6 WU) (OR per SD: 2.38, 95%CI 1.34, 4.25, and OR per SD: 2.67, 95%CI 1.26, 5.64, respectively). In receiver operating characteristic analysis, the TAPSE/PASP + PV peak velocity combined model achieved the best performance (AUC: 0.8584, sensitivity: 83.33%, specificity: 72.16%). Internal verification showed stable performance (AUC: 0.8591, sensitivity: 88.10%, specificity: 68.04%). The net benefit of this model was greater than other models when it came to a wide range probability threshold in decision curve analysis. TAPSE/PASP + PV the peak velocity model may have great value in predicting adult ASD-PAH patients with operability potential, which could help clinicians make the treatment decision for follow-up patients.

Analysis of the nonperfused volume ratio of adenomyosis from MRI images based on fewshot learning

The nonperfused volume (NPV) ratio is the key to the success of high intensity focused ultrasound (HIFU) ablation treatment of adenomyosis. However, there are no qualitative interpretation standards for predicting the NPV ratio of adenomyosis using magnetic resonance imaging (MRI) before HIFU ablation treatment, which leading to inter-reader variability. Convolutional neural networks (CNNs) have achieved state-of-the-art performance in the automatic disease diagnosis of MRI. Since the use of HIFU to treat adenomyosis is a novel treatment, there is not enough MRI data to support CNNs. We proposed a novel few-shot learning framework that extends CNNs to predict NPV ratio of HIFU ablation treatment for adenomyosis. We collected a dataset from 208 patients with adenomyosis who underwent MRI examination before and after HIFU treatment. Our proposed method was trained and evaluated by fourfold cross validation. This framework obtained sensitivity of 85.6%, 89.6% and 92.8% at 0.799, 0.980 and 1.180 FPs per patient. In the receiver operating characteristics analysis for NPV ratio of adenomyosis, our proposed method received the area under the curve of 0.8233, 0.8289, 0.8412, 0.8319, 0.7010, 0.7637, 0.8375, 0.8219, 0.8207, 0.9812 for the classifications of NPV ratio interval [0%-10%), [10%-20%), …, [90%-100%], respectively. The present study demonstrated that few-shot learning on NPV ratio prediction of HIFU ablation treatment for adenomyosis may contribute to the selection of eligible patients and the pre-judgment of clinical efficacy.

Deep convolutional neural networks for multiplanar lung nodule detection: Improvement in small nodule identification

PURPOSE

Early detection of lung cancer is of importance since it can increase patients' chances of survival. To detect nodules accurately during screening, radiologists would commonly take the axial, coronal, and sagittal planes into account, rather than solely the axial plane in clinical evaluation. Inspired by clinical work, the paper aims to develop an accurate deep learning framework for nodule detection by a combination of multiple planes.

METHODS

The nodule detection system is designed in two stages, multiplanar nodule candidate detection, multiscale false positive (FP) reduction. At the first stage, a deeply supervised encoder-decoder network is trained by axial, coronal, and sagittal slices for the candidate detection task. All possible nodule candidates from the three different planes are merged. To further refine results, a three-dimensional multiscale dense convolutional neural network that extracts multiscale contextual information is applied to remove non-nodules. In the public LIDC-IDRI dataset, 888 computed tomography scans with 1186 nodules accepted by at least three of four radiologists are selected to train and evaluate our proposed system via a tenfold cross-validation scheme. The free-response receiver operating characteristic curve is used for performance assessment.

RESULTS

The proposed system achieves a sensitivity of 94.2% with 1.0 FP/scan and a sensitivity of 96.0% with 2.0 FPs/scan. Although it is difficult to detect small nodules (i.e., <6 mm), our designed CAD system reaches a sensitivity of 93.4% (95.0%) of these small nodules at an overall FP rate of 1.0 (2.0) FPs/scan. At the nodule candidate detection stage, results show that the system with a multiplanar method is capable to detect more nodules compared to using a single plane.

CONCLUSION

Our approach achieves good performance not only for small nodules but also for large lesions on this dataset. This demonstrates the effectiveness of our developed CAD system for lung nodule detection.

A nomogram for predicting the risk of lymph node metastasis in T1-2 non-small-cell lung cancer based on PET/CT and clinical characteristics

Background

Accurately predicting the risk level for a lymph node metastasis is critical in the treatment of non-small cell lung cancer (NSCLC). This study aimed to construct a novel nomogram to identify patients with a risk of lymph node metastasis in T1–2 NSCLC based on positron emission tomography/computed tomography (PET/CT) and clinical characteristics.

Methods

From January 2011 to November 2017, the records of 318 consecutive patients who had undergone PET/CT examination within 30 days before surgical resection for clinical T1–2 NSCLC were retrospectively reviewed. A nomogram to predict the risk of lymph node metastasis was constructed. The model was confirmed using bootstrap resampling, and an independent validation cohort contained 156 patients from June 2017 to February 2020 at another institution.

Results

Six factors [age, tumor location, histology, the lymph node maximum standardized uptake value (SUVmax), the tumor SUVmax and the carcinoembryonic antigen (CEA) value] were identified and entered into the nomogram. The nomogram developed based on the analysis showed robust discrimination, with an area under the receiver operating characteristic curve of 0.858 in the primary cohort and 0.749 in the validation cohort. The calibration curve for the probability of lymph node metastasis showed excellent concordance between the predicted and actual results. Decision curve analysis suggested that the nomogram was clinically useful.

Conclusions

We set up and validated a novel and effective nomogram that can predict the risk of lymph node metastasis for individual patients with T1–2 NSCLC. This model may help clinicians to make treatment recommendations for individuals.

Automated detection and delineation of hepatocellular carcinoma on multiphasic contrast-enhanced MRI using deep learning

PURPOSE

Liver Imaging Reporting and Data System (LI-RADS) uses multiphasic contrast-enhanced imaging for hepatocellular carcinoma (HCC) diagnosis. The goal of this feasibility study was to establish a proof-of-principle concept towards automating the application of LI-RADS, using a deep learning algorithm trained to segment the liver and delineate HCCs on MRI automatically.

METHODS

In this retrospective single-center study, multiphasic contrast-enhanced MRIs using T1-weighted breath-hold sequences acquired from 2010 to 2018 were used to train a deep convolutional neural network (DCNN) with a U-Net architecture. The U-Net was trained (using 70% of all data), validated (15%) and tested (15%) on 174 patients with 231 lesions. Manual 3D segmentations of the liver and HCC were ground truth. The dice similarity coefficient (DSC) was measured between manual and DCNN methods. Postprocessing using a random forest (RF) classifier employing radiomic features and thresholding (TR) of the mean neural activation was used to reduce the average false positive rate (AFPR).

RESULTS

73 and 75% of HCCs were detected on validation and test sets, respectively, using > 0.2 DSC criterion between individual lesions and their corresponding segmentations. Validation set AFPRs were 2.81, 0.77, 0.85 for U-Net, U-Net + RF, and U-Net + TR, respectively. Combining both RF and TR with the U-Net improved the AFPR to 0.62 and 0.75 for the validation and test sets, respectively. Mean DSC between automatically detected lesions using the DCNN + RF + TR and corresponding manual segmentations was 0.64/0.68 (validation/test), and 0.91/0.91 for liver segmentations.

CONCLUSION

Our DCNN approach can segment the liver and HCCs automatically. This could enable a more workflow efficient and clinically realistic implementation of LI-RADS.

Mesenchymal and Phosphatase of Regenerating Liver-3 Status in Circulating Tumor Cells May Serve as a Crucial Prognostic Marker for Assessing Relapse or Metastasis in Postoperative Patients With Colorectal Cancer

INTRODUCTION

Circulating tumor cells (CTCs) and phosphatase of regenerating liver-3 (PRL-3) have been considered to be significant prognostic indicators in metastatic colorectal cancer (CRC). This study discusses the prognostic significance of mesenchymal CTCs with PRL-3 (M+ PRL-3+ CTCs) in postoperative patients with CRC.

METHODS

We detected CTC subtypes (including epithelial CTCs, biphenotypic epithelial/mesenchymal CTCs, and mesenchymal CTCs) and PRL-3 in CTCs from the peripheral blood samples of 156 patients. Receiver operating characteristic curve analysis, Kaplan-Meier analysis, and Cox proportional hazards regression analysis were performed to identify the prognostic value of mesenchymal CTCs with PRL-3+. Immunohistochemistry was used to detect the expression of PRL-3 in tumor tissues from some of the patients to explore the connection between CTCs and tissues.

RESULTS

All CTCs were positive in all samples, both mesenchymal CTCs and PRL-3-positive cells. The count of mesenchymal and PRL-3+ CTCs was significantly associated with recurrence, and the optimal cutoff value was 2 (area under the curve = 0.690, P < 0.001). In addition, these patients had a significantly shorter median disease-free survival than those who did not fulfill the criteria (8.5 vs 24 months, P < 0.001) according to multivariable and multinomial logistic regression. Immunohistochemistry was applied to explore the associations between PRL-3 expression and significant prognostic risk factors, including recurrence (R = 0.566; P < 0.001), and M+ PRL-3+ status in CTCs (R = 0.452; P = 0.001).

DISCUSSION

The status of M+ PRL-3+ in CTCs may serve as a crucial prognostic marker for assessing clinical outcomes in CRC.

Error-corrected estimation of a diagnostic accuracy index of a biomarker against a continuous gold standard

This article concerns evaluating the effectiveness of a continuous diagnostic biomarker against a continuous gold standard that is measured with error. Extending the work of Obuchowski (2005, 2016), Wu et al (2016) suggested an accuracy index and proposed an estimator for the index with error-prone standard when the reliability coefficient is known. Combining with additional measurements (without measurement errors) on the continuous gold standard collected from some subjects, this article proposes two adaptive estimators of the accuracy index when the reliability coefficient is unknown, and further establish the consistency and asymptotic normality of these estimators. Simulation studies are conducted to compare various estimators. Data from an intervention trial on glycemic control among children with type 1 diabetes are used to illustrate the proposed methods.

Mass Detection and Segmentation in Digital Breast Tomosynthesis Using 3D-Mask Region-Based Convolutional Neural Network: A Comparative Analysis

Digital breast tomosynthesis (DBT) is an emerging breast cancer screening and diagnostic modality that uses quasi-three-dimensional breast images to provide detailed assessments of the dense tissue within the breast. In this study, a framework of a 3D-Mask region-based convolutional neural network (3D-Mask RCNN) computer-aided diagnosis (CAD) system was developed for mass detection and segmentation with a comparative analysis of performance on patient subgroups with different clinicopathological characteristics. To this end, 364 samples of DBT data were used and separated into a training dataset (n = 201) and a testing dataset (n = 163). The detection and segmentation results were evaluated on the testing set and on subgroups of patients with different characteristics, including different age ranges, lesion sizes, histological types, lesion shapes and breast densities. The results of our 3D-Mask RCNN framework were compared with those of the 2D-Mask RCNN and Faster RCNN methods. For lesion-based mass detection, the sensitivity of 3D-Mask RCNN-based CAD was 90% with 0.8 false positives (FPs) per lesion, whereas the sensitivity of the 2D-Mask RCNN- and Faster RCNN-based CAD was 90% at 1.3 and 2.37 FPs/lesion, respectively. For breast-based mass detection, the 3D-Mask RCNN generated a sensitivity of 90% at 0.83 FPs/breast, and this framework is better than the 2D-Mask RCNN and Faster RCNN, which generated a sensitivity of 90% with 1.24 and 2.38 FPs/breast, respectively. Additionally, the 3D-Mask RCNN achieved significantly (p < 0.05) better performance than the 2D methods on subgroups of samples with characteristics of ages ranged from 40 to 49 years, malignant tumors, spiculate and irregular masses and dense breast, respectively. Lesion segmentation using the 3D-Mask RCNN achieved an average precision (AP) of 0.934 and a false negative rate (FNR) of 0.053, which are better than those achieved by the 2D methods. The results suggest that the 3D-Mask RCNN CAD framework has advantages over 2D-based mass detection on both the whole data and subgroups with different characteristics.

Deep learning-based pulmonary nodule detection: Effect of slab thickness in maximum intensity projections at the nodule candidate detection stage

BACKGROUND AND OBJECTIVE

To investigate the effect of the slab thickness in maximum intensity projections (MIPs) on the candidate detection performance of a deep learning-based computer-aided detection (DL-CAD) system for pulmonary nodule detection in CT scans.

METHODS

The public LUNA16 dataset includes 888 CT scans with 1186 nodules annotated by four radiologists. From those scans, MIP images were reconstructed with slab thicknesses of 5 to 50 mm (at 5 mm intervals) and 3 to 13 mm (at 2 mm intervals). The architecture in the nodule candidate detection part of the DL-CAD system was trained separately using MIP images with various slab thicknesses. Based on ten-fold cross-validation, the sensitivity and the F₂ score were determined to evaluate the performance of using each slab thickness at the nodule candidate detection stage. The free-response receiver operating characteristic (FROC) curve was used to assess the performance of the whole DL-CAD system that took the results combined from 16 MIP slab thickness settings.

RESULTS

At the nodule candidate detection stage, the combination of results from 16 MIP slab thickness settings showed a high sensitivity of 98.0% with 46 false positives (FPs) per scan. Regarding a single MIP slab thickness of 10 mm, the highest sensitivity of 90.0% with 8 FPs/scan was reached before false positive reduction. The sensitivity increased (82.8% to 90.0%) for slab thickness of 1 to 10 mm and decreased (88.7% to 76.6%) for slab thickness of 15-50 mm. The number of FPs was decreasing with increasing slab thickness, but was stable at 5 FPs/scan at a slab thickness of 30 mm or more. After false positive reduction, the DL-CAD system, utilizing 16 MIP slab thickness settings, had the sensitivity of 94.4% with 1 FP/scan.

CONCLUSIONS

The utilization of multi-MIP images could improve the performance at the nodule candidate detection stage, even for the whole DL-CAD system. For a single slab thickness of 10 mm, the highest sensitivity for pulmonary nodule detection was reached at the nodule candidate detection stage, similar to the slab thickness usually applied by radiologists.

The role of CD28 in the prognosis of young lung adenocarcinoma patients

Background

The prognosis of lung cancer was found to be associated with a series of biomarkers related to the tumor immune microenvironment (TIME), which can modulate the biological behaviors and consequent outcomes of lung cancer. Therefore, establishing a prognostic model based on the TIME for lung cancer patients, especially young patients with lung adenocarcinoma (LUAD), is urgently needed.

Methods

In all, 809 lung cancer patients from the TCGA database and 71 young patients with LUAD in our center were involved in this study. Univariate and multivariate analysis based on clinical characteristics and TIME-related expression patterns (as evaluated by IHC) were performed to estimate prognosis and were verified by prognostic nomograms.

Results

Both LUAD and lung cancer patients with high CD28 expression had shorter disease-free survival (DFS) (P = 0.0011; P = 0.0001) but longer overall survival (OS) (P = 0.0001; P = 0.0282). TIME-related molecules combined with clinical information and genomic signatures could predict the prognosis of young patients with LUAD with robust efficiency and could be verified by the established nomogram based on the Cox regression model. In addition, CD28 expression was correlated with an abundance of lymphocytes and could modulate the TIME. Higher CD28 levels were observed in primary tumors than in metastatic tissues.

Conclusion

TIME-related molecules were identified as compelling biomarkers for predicting the prognosis of lung cancer, especially in a cohort of young patients. Furthermore, CD28, which is associated with poor DFS but long OS, might participate in the modulation of the TIME and has a different role in the prognosis of young patients with LUAD.

Subunits of ARID1 serve as novel biomarkers for the sensitivity to immune checkpoint inhibitors and prognosis of advanced non-small cell lung cancer

Introduction

Patients with advanced non-small cell lung cancer (NSCLC) benefit from treatment with immune checkpoint inhibitors (ICIs). Biomarkers such as programmed death-ligand 1 (PD-L1), the tumor mutational burden (TMB) and the mismatch repair (MMR) status are used to predict the prognosis of ICIs therapy. Nevertheless, novel biomarkers need to be further investigated, and a systematic prognostic model is needed for the evaluation of the survival risks of ICIs treatment.

Methods

A cohort of 240 patients who received ICIs from the cBioPortal for Cancer Genomics was evaluated in this research. Clinical information and targeted sequencing data were acquired for analyses. The Kaplan-Meier plot method was used to perform survival analyses, and selected variables were then confirmed by a novel nomogram constructed by the “rms” package of R software.

Results

Seven percent of the NSCLC patients harbored ARID1A mutations, while 4% of the NSCLC patients harbored ARID1B mutations. Mutations in ARID1A and ARID1B were confirmed to be associated with sensitivity to ICIs. Patients harboring these mutations were found to have a better response to treatment (ARID1A: P = 0.045; ARID1B: P = 0.034) and prolonged progression-free survival (ARID1B: P = 0.032). Here, a novel nomogram was constructed to predict the prognosis of ICIs treatment. Elevation of the TMB, enhanced expression of PD-L1 and activation of the antigen presentation process and cellular immunity were found to be correlated with ARID1A and ARID1B mutations.

Conclusion

ARID1A and ARID1B could serve as novel biomarkers for the prognosis and sensitivity to ICIs of advanced NSCLC.

Primary Categorizing and Masking Cerebral Small Vessel Disease Based on "Deep Learning System"

Objective

To supply the attending doctor’s diagnosis of the persisting of cerebral small vessel disease and speed up their work effectively, we developed a “deep learning system (DLS)” for cerebral small vessel disease predication. The reliability and the disease area segmentation accuracy, of the proposed DLS, was also investigated.

Methods

A deep learning model based on the convolutional neural network was designed and trained on 1,010 DWI b1000 images from 1010 patients diagnosed with segmentation of subcortical infarction, 359 T2^∗ images from 359 patients diagnosed with segmentation of cerebral microbleed, as well as 824 T1-weighted and T2-FLAIR images from 824 patients diagnosed with segmentation of lacune and WMH. Dicw accuracy, recall, and f1-score were calculated to evaluate the proposed deep learning model. Finally, we also compared the DLS prediction capability with that of 6 doctors with 3 to 18 years’ clinical experience (8 ± 6 years).

Results

The results support that an appropriately trained DLS can achieve a high-level dice accuracy, 0.598 in the training section over all these four classifications on 30 patients (0.576 for young neuroradiologists), validation accuracy is 0.496 in lacune, 0.666 in WMH, 0.728 in subcortical infarction, and 0.503 in cerebral microbleeds. It is comparable to attending doctor with a few years of experience, regardless of whether the emphasis is placed on the segmentation or detection of lesions with less time-spending compared with manual analysis, about 4.4 s/case, which is dramatically less than doctors about 634 s/case.

Conclusion

The results of our comparison lend support to the case that an appropriately trained DLS can be trusted to the same extent as one would trust an attending doctor with a few years of experience, regardless of whether the emphasis is placed on the segmentation or detection of lesions.

Fully automated intracranial aneurysm detection and segmentation from digital subtraction angiography series using an end-to-end spatiotemporal deep neural network

Background

Intracranial aneurysms (IAs) are common in the population and may cause death.

Objective

To develop a new fully automated detection and segmentation deep neural network based framework to assist neurologists in evaluating and contouring intracranial aneurysms from 2D+time digital subtraction angiography (DSA) sequences during diagnosis.

Methods

The network structure is based on a general U-shaped design for medical image segmentation and detection. The network includes a fully convolutional technique to detect aneurysms in high-resolution DSA frames. In addition, a bidirectional convolutional long short-term memory module is introduced at each level of the network to capture the change in contrast medium flow across the 2D DSA frames. The resulting network incorporates both spatial and temporal information from DSA sequences and can be trained end-to-end. Furthermore, deep supervision was implemented to help the network converge. The proposed network structure was trained with 2269 DSA sequences from 347 patients with IAs. After that, the system was evaluated on a blind test set with 947 DSA sequences from 146 patients.

Results

Of the 354 aneurysms, 316 (89.3%) were successfully detected, corresponding to a patient level sensitivity of 97.7% at an average false positive number of 3.77 per sequence. The system runs for less than one second per sequence with an average dice coefficient score of 0.533.

Conclusions

This deep neural network assists in successfully detecting and segmenting aneurysms from 2D DSA sequences, and can be used in clinical practice.

Computer aided detection for laterally spreading tumors and sessile serrated adenomas during colonoscopy

Background

Evidence has shown that deep learning computer aided detection (CADe) system achieved high overall detection accuracy for polyp detection during colonoscopy.

Aim

The detection performance of CADe system on non-polypoid laterally spreading tumors (LSTs) and sessile serrated adenomas/polyps (SSA/Ps), with higher risk for malignancy transformation and miss rate, has not been exclusively investigated.

Methods

A previously validated deep learning CADe system for polyp detection was tested exclusively on LSTs and SSA/Ps. 1451 LST images from 184 patients were collected between July 2015 and January 2019, 82 SSA/Ps videos from 26 patients were collected between September 2018 and January 2019. The per-frame sensitivity and per-lesion sensitivity were calculated.

Results

(1) For LSTs image dataset, the system achieved an overall per-image sensitivity and per-lesion sensitivity of 94.07% (1365/1451) and 98.99% (197/199) respectively. The per-frame sensitivity for LST-G(H), LST-G(M), LST-NG(F), LST-NG(PD) was 93.97% (343/365), 98.72% (692/701), 85.71% (324/378) and 85.71% (6/7) respectively. The per-lesion sensitivity of each subgroup was 100.00% (71/71), 100.00% (64/64), 98.31% (58/59) and 80.00% (4/5). (2) For SSA/Ps video dataset, the system achieved an overall per-frame sensitivity and per-lesion sensitivity of 84.10% (15883/18885) and 100.00% (42/42), respectively.

Conclusions

This study demonstrated that a local-feature-prioritized automatic CADe system could detect LSTs and SSA/Ps with high sensitivity. The per-frame sensitivity for non-granular LSTs and small SSA/Ps should be further improved.