Deep learning-based survival analysis for brain metastasis patients with the national cancer database

Explainable deep learning-based survival prediction for non-small cell lung cancer patients undergoing radical radiotherapy

BACKGROUND AND PURPOSE

Survival is frequently assessed using Cox proportional hazards (CPH) regression; however, CPH may be too simplistic as it assumes a linear relationship between covariables and the outcome. Alternative, non-linear machine learning (ML)-based approaches, such as random survival forests (RSFs) and, more recently, deep learning (DL) have been proposed; however, these techniques are largely black-box in nature, limiting explainability. We compared CPH, RSF and DL to predict overall survival (OS) of non-small cell lung cancer (NSCLC) patients receiving radiotherapy using pre-treatment covariables. We employed explainable techniques to provide insights into the contribution of each covariable on OS prediction.

MATERIALS AND METHODS

The dataset contained 471 stage I-IV NSCLC patients treated with radiotherapy. We built CPH, RSF and DL OS prediction models using several baseline covariable combinations. 10-fold Monte-Carlo cross-validation was employed with a split of 70%:10%:20% for training, validation and testing, respectively. We primarily evaluated performance using the concordance index (C-index) and integrated Brier score (IBS). Local interpretable model-agnostic explanation (LIME) values, adapted for use in survival analysis, were computed for each model.

RESULTS

The DL method exhibited a significantly improved C-index of 0.670 compared to the CPH and a significantly improved IBS of 0.121 compared to the CPH and RSF approaches. LIME values suggested that, for the DL method, the three most important covariables in OS prediction were stage, administration of chemotherapy and oesophageal mean radiation dose.

CONCLUSION

We show that, using pre-treatment covariables, a DL approach demonstrates superior performance over CPH and RSF for OS prediction and use explainable techniques to provide transparency and interpretability.

Use of survival support vector machine combined with random survival forest to predict the survival of nasopharyngeal carcinoma patients

Background

The Cox regression model is not sufficiently accurate to predict the survival prognosis of nasopharyngeal carcinoma (NPC) patients. It is impossible to calculate and rank the importance of impact factors due to the low predictive accuracy of the Cox regression model. So, we developed a system. Using the SEER (The Surveillance, Epidemiology, and End Results) database data on NPC patients, we proposed the use of random survival forest (RSF) and survival-support vector machine (SVM) from the machine learning methods to develop a survival prediction system specifically for NPC patients. This approach aimed to make up for the insufficiency of the Cox regression model. We also used the Cox regression model to validate the development of the nomogram and compared it with machine learning methods.

Methods

A total of 1,683 NPC patients were extracted from the SEER database from January 2010 to December 2015. We used R language for modeling work, established the nomogram of survival prognosis of NPC patients by Cox regression model, ranked the correlation of influencing factors by RSF model VIMP (variable important) method, developed a survival prognosis system for NPC patients based on survival-SVM, and used C-index for model evaluation and performance comparison.

Results

Although the Cox regression models can be developed to predict the prognosis of NPC patients, their accuracy was lower than that of machine learning methods. When we substituted the data for the Cox model, the C-index for the training set was only 0.740, and the C-index for the test set was 0.721. In contrast, the C index of the survival-SVM model was 0.785. The C-index of the RSF model was 0.729. The importance ranking of each variable could be obtained according to the VIMP method.

Conclusions

The prediction results from the Cox model are not as good as those of the RSF method and survival-SVM based on the machine learning method. For the survival prognosis of NPC patients, the machine learning method can be considered for clinical application.

Predicting peritumoral edema development after gamma knife radiosurgery of meningiomas using machine learning methods: a multicenter study

OBJECTIVES

Edema is a complication of gamma knife radiosurgery (GKS) in meningioma patients that leads to a variety of consequences. The aim of this study is to construct radiomics-based machine learning models to predict post-GKS edema development.

METHODS

In total, 445 meningioma patients who underwent GKS in our institution were enrolled and partitioned into training and internal validation datasets (8:2). A total of 150 cases from multicenter data were included as the external validation dataset. In each case, 1132 radiomics features were extracted from each pre-treatment MRI sequence (contrast-enhanced T1WI, T2WI, and ADC maps). Nine clinical features and eight semantic features were also generated. Nineteen random survival forest (RSF) and nineteen neural network (DeepSurv) models with different combinations of radiomics, clinical, and semantic features were developed with the training dataset, and evaluated with internal and external validation. A nomogram was derived from the model achieving the highest C-index in external validation.

RESULTS

All the models were successfully validated on both validation datasets. The RSF model incorporating clinical, semantic, and ADC radiomics features achieved the best performance with a C-index of 0.861 (95% CI: 0.748-0.975) in internal validation, and 0.780 (95% CI: 0.673-0.887) in external validation. It stratifies high-risk and low-risk cases effectively. The nomogram based on the predicted risks provided personalized prediction with a C-index of 0.962 (95%CI: 0.951-0.973) and satisfactory calibration.

CONCLUSION

This RSF model with a nomogram could represent a non-invasive and cost-effective tool to predict post-GKS edema risk, thus facilitating personalized decision-making in meningioma treatment.

CLINICAL RELEVANCE STATEMENT

The RSF model with a nomogram built in this study represents a handy, non-invasive, and cost-effective tool for meningioma patients to assist in better counselling on the risks, appropriate individual treatment decisions, and customized follow-up plans.

KEY POINTS

• Machine learning models were built to predict post-GKS edema in meningioma. The random survival forest model with clinical, semantic, and ADC radiomics features achieved excellent performance. • The nomogram based on the predicted risks provides personalized prediction with a C-index of 0.962 (95%CI: 0.951-0.973) and satisfactory calibration and shows the potential to assist in better counselling, appropriate treatment decisions, and customized follow-up plans. • Given the excellent performance and convenient acquisition of the conventional sequence, we envision that this non-invasive and cost-effective tool will facilitate personalized medicine in meningioma treatment.

Deep survival analysis using pseudo values and its application to predict the recurrence of stage IV colorectal cancer after tumor resection

An improved DeepSurv model is proposed for predicting the prognosis of colorectal cancer patients at stage IV. Our model, called as PseudoDeepSurv, is optimized by a novel loss function, which is the combination of the average negative log partial likelihood and the mean-squared error derived from the pseudo-observations approach. The public BioStudies dataset including 999 patients was utilized for performance evaluation. Our PseudoDeepSurv model produced a C-index of 0.684 and 0.633 on the training and testing dataset, respectively. While for the original DeepSurv model, the corresponding values are 0.671 and 0.618, respectively.

A novel staging system based on deep learning for overall survival in patients with esophageal squamous cell carcinoma

PURPOSE

We developed DeepSurv, a deep learning approach for predicting overall survival (OS) in patients with esophageal squamous cell carcinoma (ESCC). We validated and visualized the novel staging system based on DeepSurv using data from multiple cohorts.

METHODS

Totally 6020 ESCC patients diagnosed from January 2010 to December 2018 were included in the present study from the Surveillance, Epidemiology, and End Results database (SEER), randomly assigned to the training and test cohorts. We developed, validated and visualized a deep learning model that included 16 prognostic factors; then a novel staging system was further constructed based on the total risk score derived from the deep learning model. The classification performance at 3-year and 5-year OS was assessed by the receiver-operating characteristic (ROC) curve. Calibration curve and the Harrell's concordance index (C-index) were also used to comprehensively assess the predictive performance of the deep learning model. Decision curve analysis (DCA) was utilized to assess the clinical utility of the novel staging system.

RESULTS

A more applicable and accurate deep learning model was established, which outperformed the traditional nomogram in predicting OS in the test cohort (C-index: 0.732 [95% CI 0.714-0.750] versus 0.671 [95% CI 0.647-0.695]). The ROC curves at 3-year and 5-year OS for the model also showed good discrimination ability in the test cohort (Area Under the Curve [AUC] at 3-/5-year OS = 0.805/0.825). Moreover, using our novel staging system, we observed a clear survival difference among different risk groups (P < 0.001) and a significant positive net benefit in the DCA.

CONCLUSIONS

A novel deep learning-based staging system was constructed for patients with ESCC, which performed a significant discriminability for survival probability. Moreover, an easy-to-use web-based tool based on the deep learning model was also implemented, offering convenience for personalized survival prediction. We developed a deep learning-based system that stages patients with ESCC according to their survival probability. We also created a web-based tool that uses this system to predict individual survival outcomes.

Deep Learning-Based Time-to-Death Prediction Model for COVID-19 Patients Using Clinical Data and Chest Radiographs

Accurate estimation of mortality and time to death at admission for COVID-19 patients is important and several deep learning models have been created for this task. However, there are currently no prognostic models which use end-to-end deep learning to predict time to event for admitted COVID-19 patients using chest radiographs and clinical data. We retrospectively implemented a new artificial intelligence model combining DeepSurv (a multiple-perceptron implementation of the Cox proportional hazards model) and a convolutional neural network (CNN) using 1356 COVID-19 inpatients. For comparison, we also prepared DeepSurv only with clinical data, DeepSurv only with images (CNNSurv), and Cox proportional hazards models. Clinical data and chest radiographs at admission were used to estimate patient outcome (death or discharge) and duration to the outcome. The Harrel's concordance index (c-index) of the DeepSurv with CNN model was 0.82 (0.75-0.88) and this was significantly higher than the DeepSurv only with clinical data model (c-index = 0.77 (0.69-0.84), p = 0.011), CNNSurv (c-index = 0.70 (0.63-0.79), p = 0.001), and the Cox proportional hazards model (c-index = 0.71 (0.63-0.79), p = 0.001). These results suggest that the time-to-event prognosis model became more accurate when chest radiographs and clinical data were used together.

A novel risk classification system based on the eighth edition of TNM frameworks for esophageal adenocarcinoma patients: A deep learning approach

Objective

To develop and validate a deep learning predictive model with better performance in survival estimation of esophageal adenocarcinoma (EAC).

Method

Cases diagnosed between January 2010 and December 2018 were extracted from the Surveillance, Epidemiology, and End Results (SEER) database. A deep learning survival neural network was developed and validated based on 17 variables, including demographic information, clinicopathological characteristics, and treatment details. Based on the total risk score derived from this algorithm, a novel risk classification system was constructed and compared with the 8th edition of the tumor, node, and metastasis (TNM) staging system.

Results

Of 7,764 EAC patients eligible for the study, 6,818 (87.8%) were men and the median (interquartile range, IQR) age was 65 (58-72) years. The deep learning model generated significantly superior predictions to the 8th edition staging system on the test data set (C-index: 0.773 [95% CI, 0.757-0.789] vs. 0.683 [95% CI, 0.667-0.699]; P < 0.001). Calibration curves revealed that the deep learning model was well calibrated for 1- and 3-year OS, most points almost directly distributing on the 45° line. Decision curve analyses (DCAs) showed that the novel risk classification system exhibited a more significant positive net benefit than the TNM staging system. A user-friendly and precise web-based calculator with a portably executable file was implemented to visualize the deep learning predictive model.

Conclusion

A deep learning predictive model was developed and validated, which possesses more excellent calibration and discrimination abilities in survival prediction of EAC. The novel risk classification system based on the deep learning algorithm may serve as a useful tool in clinical decision making given its easy-to-use and better clinical applicability.

Predicting Survival in Patients with Brain Tumors: Current State-of-the-Art of AI Methods Applied to MRI

Given growing clinical needs, in recent years Artificial Intelligence (AI) techniques have increasingly been used to define the best approaches for survival assessment and prediction in patients with brain tumors. Advances in computational resources, and the collection of (mainly) public databases, have promoted this rapid development. This narrative review of the current state-of-the-art aimed to survey current applications of AI in predicting survival in patients with brain tumors, with a focus on Magnetic Resonance Imaging (MRI). An extensive search was performed on PubMed and Google Scholar using a Boolean research query based on MeSH terms and restricting the search to the period between 2012 and 2022. Fifty studies were selected, mainly based on Machine Learning (ML), Deep Learning (DL), radiomics-based methods, and methods that exploit traditional imaging techniques for survival assessment. In addition, we focused on two distinct tasks related to survival assessment: the first on the classification of subjects into survival classes (short and long-term or eventually short, mid and long-term) to stratify patients in distinct groups. The second focused on quantification, in days or months, of the individual survival interval. Our survey showed excellent state-of-the-art methods for the first, with accuracy up to ∼98%. The latter task appears to be the most challenging, but state-of-the-art techniques showed promising results, albeit with limitations, with C-Index up to ∼0.91. In conclusion, according to the specific task, the available computational methods perform differently, and the choice of the best one to use is non-univocal and dependent on many aspects. Unequivocally, the use of features derived from quantitative imaging has been shown to be advantageous for AI applications, including survival prediction. This evidence from the literature motivates further research in the field of AI-powered methods for survival prediction in patients with brain tumors, in particular, using the wealth of information provided by quantitative MRI techniques.

An automated approach for predicting glioma grade and survival of LGG patients using CNN and radiomics

Objectives

To develop and validate an efficient and automatically computational approach for stratifying glioma grades and predicting survival of lower-grade glioma (LGG) patients using an integration of state-of-the-art convolutional neural network (CNN) and radiomics.

Method

This retrospective study reviewed 470 preoperative MR images of glioma from BraTs public dataset (n=269) and Jinling hospital (n=201). A fully automated pipeline incorporating tumor segmentation and grading was developed, which can avoid variability and subjectivity of manual segmentations. First, an integrated approach by fusing CNN features and radiomics features was employed to stratify glioma grades. Then, a deep-radiomics signature based on the integrated approach for predicting survival of LGG patients was developed and subsequently validated in an independent cohort.

Results

The performance of tumor segmentation achieved a Dice coefficient of 0.81. The intraclass correlation coefficients (ICCs) of the radiomics features between the segmentation network and physicians were all over 0.75. The performance of glioma grading based on integrated approach achieved the area under the curve (AUC) of 0.958, showing the effectiveness of the integrated approach. The multivariable Cox regression results demonstrated that the deep-radiomics signature remained an independent prognostic factor and the integrated nomogram showed significantly better performance than the clinical nomogram in predicting overall survival of LGG patients (C-index: 0.865 vs. 0.796, P=0.005).

Conclusion

The proposed integrated approach can be noninvasively and efficiently applied in prediction of gliomas grade and survival. Moreover, our fully automated pipeline successfully achieved computerized segmentation instead of manual segmentation, which shows the potential to be a reproducible approach in clinical practice.

An energy-efficient in-memory computing architecture for survival data analysis based on resistive switching memories

One of the objectives fostered in medical science is the so-called precision medicine, which requires the analysis of a large amount of survival data from patients to deeply understand treatment options. Tools like machine learning (ML) and deep neural networks are becoming a de-facto standard. Nowadays, computing facilities based on the Von Neumann architecture are devoted to these tasks, yet rapidly hitting a bottleneck in performance and energy efficiency. The in-memory computing (IMC) architecture emerged as a revolutionary approach to overcome that issue. In this work, we propose an IMC architecture based on resistive switching memory (RRAM) crossbar arrays to provide a convenient primitive for matrix-vector multiplication in a single computational step. This opens massive performance improvement in the acceleration of a neural network that is frequently used in survival analysis of biomedical records, namely the DeepSurv. We explored how the synaptic weights mapping strategy and the programming algorithms developed to counter RRAM non-idealities expose a performance/energy trade-off. Finally, we discussed how this application is tailored for the IMC architecture rather than being executed on commodity systems.

Applying Artificial Neural Networks to Develop a Decision Support Tool for Tis-4N0M0 Non-Small-Cell Lung Cancer Treated With Stereotactic Body Radiotherapy

PURPOSE

Clear evidence indicating whether surgery or stereotactic body radiation therapy (SBRT) is best for non-small-cell lung cancer (NSCLC) is lacking. SBRT has many advantages. We used artificial neural networks (NNs) to predict treatment outcomes for patients with NSCLC receiving SBRT, aiming to aid in decision making.

PATIENTS AND METHODS

Among consecutive patients receiving SBRT between 2005 and 2019 in our institution, we retrospectively identified those with Tis-T4N0M0 NSCLC. We constructed two NNs for prediction of overall survival (OS) and cancer progression in the first 5 years after SBRT, which were tested using an internal and an external test data set. We performed risk group stratification, wherein 5-year OS and cancer progression were stratified into three groups.

RESULTS

In total, 692 patients in our institution and 100 patients randomly chosen in the external institution were enrolled. The NNs resulted in concordance indexes for OS of 0.76 (95% CI, 0.73 to 0.79), 0.68 (95% CI, 0.60 to 0.75), and 0.69 (95% CI, 0.61 to 0.76) and area under the curve for cancer progression of 0.80 (95% CI, 0.75 to 0.84), 0.72 (95% CI, 0.60 to 0.83), and 0.70 (95% CI, 0.57 to 0.81) in the training, internal test, and external test data sets, respectively. The survival and cumulative incidence curves were significantly stratified. NNs selected low-risk cancer progression groups of 5.6%, 6.9%, and 7.0% in the training, internal test, and external test data sets, respectively, suggesting that 48% of patients with peripheral Tis-4N0M0 NSCLC can be at low-risk for cancer progression.

CONCLUSION

Predictions of SBRT outcomes using NNs were useful for Tis-4N0M0 NSCLC. Our results are anticipated to open new avenues for NN predictions and provide decision-making guidance for patients and physicians.

AutoBRB: An automated belief rule base model for pathologic complete response prediction in gastric cancer

Gastric cancer is one of the most severe malignant lesions. Neoadjuvant chemotherapy (NAC) has proven to be an effective method in gastric cancer treatment, and patients who achieved the pathologic complete response (pCR) after NAC can improve survival time further. To accurately predict pCR in an interpretable way, a new automated belief rule base (AutoBRB) model is developed with careful data analysis in this paper. In AutoBRB, to determine the referential values that are important for the rule building, both the information gain ratio and expert knowledge are used, while a table-based strategy is designed to initialize the belief degrees for each rule. Then, the differential evolution (DE) algorithm is employed and modified for model optimization to improve the model's performance. Finally, with the help of training data, an adaptive searching strategy is designed to set the confidence threshold for the final prediction. The experimental results demonstrate that AutoBRB shows a more reasonable performance on the prediction of pCR.

Deep Learning for Outcome Prediction in Neurosurgery: A Systematic Review of Design, Reporting, and Reproducibility

Deep learning (DL) is a powerful machine learning technique that has increasingly been used to predict surgical outcomes. However, the large quantity of data required and lack of model interpretability represent substantial barriers to the validity and reproducibility of DL models. The objective of this study was to systematically review the characteristics of DL studies involving neurosurgical outcome prediction and to assess their bias and reporting quality. Literature search using the PubMed, Scopus, and Embase databases identified 1949 records of which 35 studies were included. Of these, 32 (91%) developed and validated a DL model while 3 (9%) validated a pre-existing model. The most commonly represented subspecialty areas were oncology (16 of 35, 46%), spine (8 of 35, 23%), and vascular (6 of 35, 17%). Risk of bias was low in 18 studies (51%), unclear in 5 (14%), and high in 12 (34%), most commonly because of data quality deficiencies. Adherence to transparent reporting of a multivariable prediction model for individual prognosis or diagnosis reporting standards was low, with a median of 12 transparent reporting of a multivariable prediction model for individual prognosis or diagnosis items (39%) per study not reported. Model transparency was severely limited because code was provided in only 3 studies (9%) and final models in 2 (6%). With the exception of public databases, no study data sets were readily available. No studies described DL models as ready for clinical use. The use of DL for neurosurgical outcome prediction remains nascent. Lack of appropriate data sets poses a major concern for bias. Although studies have demonstrated promising results, greater transparency in model development and reporting is needed to facilitate reproducibility and validation.

Prognostic risk stratification of gliomas using deep learning in digital pathology images

Background

Evaluation of tumor-tissue images stained with hematoxylin and eosin (H&E) is pivotal in diagnosis, yet only a fraction of the rich phenotypic information is considered for clinical care. Here, we propose a survival deep learning (SDL) framework to extract this information to predict glioma survival.

Methods

Digitized whole slide images were downloaded from The Cancer Genome Atlas (TCGA) for 766 diffuse glioma patients, including isocitrate dehydrogenase (IDH)-mutant/1p19q-codeleted oligodendroglioma, IDH-mutant/1p19q-intact astrocytoma, and IDH-wildtype astrocytoma/glioblastoma. Our SDL framework employs a residual convolutional neural network with a survival model to predict patient risk from H&E-stained whole-slide images. We used statistical sampling techniques and randomized the transformation of images to address challenges in learning from histology images. The SDL risk score was evaluated in traditional and recursive partitioning (RPA) survival models.

Results

The SDL risk score demonstrated substantial univariate prognostic power (median concordance index of 0.79 [se: 0.01]). After adjusting for age and World Health Organization 2016 subtype, the SDL risk score was significantly associated with overall survival (OS; hazard ratio = 2.45; 95% CI: 2.01 to 3.00). Four distinct survival risk groups were characterized by RPA based on SDL risk score, IDH status, and age with markedly different median OS ranging from 1.03 years to 14.14 years.

Conclusions

The present study highlights the independent prognostic power of the SDL risk score for objective and accurate prediction of glioma outcomes. Further, we show that the RPA delineation of patient-specific risk scores and clinical prognostic factors can successfully demarcate the OS of glioma patients.

Machine Learning Incorporating Host Factors for Predicting Survival in Head and Neck Squamous Cell Carcinoma Patients

Prognostication for cancer patients is integral for patient counseling and treatment planning, yet providing accurate prediction can be challenging using existing patient-specific clinical indicators and host factors. In this work, we evaluated common machine learning models in predicting head and neck squamous cell carcinoma (HNSCC) patients' overall survival based on demographic, clinical features and host factors. We found random survival forest had best performance among the models evaluated, which achieved a C-index of 0.729 and AUROC of 0.792 in predicting two-year overall survival. In addition, we verified that host factors are independently predictive of HNSCC overall survival, which improved the C-index by a margin of 0.026 and the AUROC by 0.034. Due to the strong correlation among host factors, we showed that proper dimension reduction is an important step before their incorporation into the machine learning models, which provides a host factor score reflecting the patients' nutrition and inflammation status. The score by itself showed excellent discriminating capacity with the high-risk group having a hazard ratio of 3.76 (1.93-7.32, p < 0.0001) over the low-risk group. The hazard ratios were further improved to 7.41 (3.66-14.98, p < 0.0001) by the random survival forest model after including demographic and clinical features.

Machine learning in oncology—Perspectives in patient-reported outcome research

Background

Increasing data volumes in oncology pose new challenges for data analysis. Machine learning, a branch of artificial intelligence, can identify patterns even in very large and less structured datasets.

Objective

This article provides an overview of the possible applications for machine learning in oncology. Furthermore, the potential of machine learning in patient-reported outcome (PRO) research is discussed.

Materials and methods

We conducted a selective literature search (PubMed, MEDLINE, IEEE Xplore) and discuss current research.

Results

There are three primary applications for machine learning in oncology: (1) cancer detection or classification; (2) overall survival prediction or risk assessment; and (3) supporting therapy decision-making and prediction of treatment response. Generally, machine learning approaches in oncology PRO research are scarce and few studies integrate PRO data into machine learning models.

Discussion

Machine learning is a promising area of oncology, but few models have been transferred into clinical practice. The promise of personalized cancer therapy and shared decision-making through machine learning has yet to be realized. As an equally important emerging research area in oncology, PROs should also be incorporated into machine learning approaches. To gather the data necessary for this, broad implementation of PRO assessments in clinical practice, as well as the harmonization of existing datasets, is suggested.