A denoised multi-omics integration framework for cancer subtype classification and survival prediction.

The availability of high-throughput sequencing data creates opportunities to comprehensively understand human diseases as well as challenges to train machine learning models using such high dimensions of data. Here, we propose a denoised multi-omics integration framework, which contains a distribution-based feature denoising algorithm, Feature Selection with Distribution (FSD), for dimension reduction and a multi-omics integration framework, Attention Multi-Omics Integration (AttentionMOI) to predict cancer prognosis and identify cancer subtypes. We demonstrated that FSD improved model performance either using single omic data or multi-omics data in 15 The Cancer Genome Atlas Program (TCGA) cancers for survival prediction and kidney cancer subtype identification. And our integration framework AttentionMOI outperformed machine learning models and current multi-omics integration algorithms with high dimensions of features. Furthermore, FSD identified features that were associated to cancer prognosis and could be considered as biomarkers.

SurvConvMixer: robust and interpretable cancer survival prediction based on ConvMixer using pathway-level gene expression images.

Cancer is one of the leading causes of deaths worldwide. Survival analysis and prediction of cancer patients is of great significance for their precision medicine. The robustness and interpretability of the survival prediction models are important, where robustness tells whether a model has learned the knowledge, and interpretability means if a model can show human what it has learned. In this paper, we propose a robust and interpretable model SurvConvMixer, which uses pathways customized gene expression images and ConvMixer for cancer short-term, mid-term and long-term overall survival prediction. With ConvMixer, the representation of each pathway can be learned respectively. We show the robustness of our model by testing the trained model on absolutely untrained external datasets. The interpretability of SurvConvMixer depends on gradient-weighted class activation mapping (Grad-Cam), by which we can obtain the pathway-level activation heat map. Then wilcoxon rank-sum tests are conducted to obtain the statistically significant pathways, thereby revealing which pathways the model focuses on more. SurvConvMixer achieves remarkable performance on the short-term, mid-term and long-term overall survival of lung adenocarcinoma, lung squamous cell carcinoma and skin cutaneous melanoma, and the external validation tests show that SurvConvMixer can generalize to external datasets so that it is robust. Finally, we investigate the activation maps generated by Grad-Cam, after wilcoxon rank-sum test and Kaplan-Meier estimation, we find that some survival-related pathways play important role in SurvConvMixer.

Indirect functional connectivity does not predict overall survival in glioblastoma.

BACKGROUND: Lesion network mapping (LNM) is a popular framework to assess clinical syndromes following brain injury. The classical approach involves embedding lesions from patients into a normative functional connectome and using the corresponding functional maps as proxies for disconnections. However, previous studies indicated limited predictive power of this approach in behavioral deficits. We hypothesized similarly low predictiveness for overall survival (OS) in glioblastoma (GBM). METHODS: A retrospective dataset of patients with GBM was included (n = 99). Lesion masks were registered in the normative space to compute disconnectivity maps. The brain functional normative connectome consisted in data from 173 healthy subjects obtained from the Human Connectome Project. A modified version of the LNM was then applied to core regions of GBM masks. Linear regression, classification, and principal component (PCA) analyses were conducted to explore the relationship between disconnectivity and OS. OS was considered both as continuous and categorical (low, intermediate, and high survival) variable. RESULTS: The results revealed no significant associations between OS and network disconnection strength when analyzed at both voxel-wise and classification levels. Moreover, patients stratified into different OS groups did not exhibit significant differences in network connectivity patterns. The spatial similarity among the first PCA of network maps for each OS group suggested a lack of distinctive network patterns associated with survival duration. CONCLUSIONS: Compared with indirect structural measures, functional indirect mapping does not provide significant predictive power for OS in patients with GBM. These findings are consistent with previous research that demonstrated the limitations of indirect functional measures in predicting clinical outcomes, underscoring the need for more comprehensive methodologies and a deeper understanding of the factors influencing clinical outcomes in this challenging disease.

Penalized deep partially linear cox models with application to CT scans of lung cancer patients.

Lung cancer is a leading cause of cancer mortality globally, highlighting the importance of understanding its mortality risks to design effective patient-centered therapies. The National Lung Screening Trial (NLST) employed computed tomography texture analysis, which provides objective measurements of texture patterns on CT scans, to quantify the mortality risks of lung cancer patients. Partially linear Cox models have gained popularity for survival analysis by dissecting the hazard function into parametric and nonparametric components, allowing for the effective incorporation of both well-established risk factors (such as age and clinical variables) and emerging risk factors (eg, image features) within a unified framework. However, when the dimension of parametric components exceeds the sample size, the task of model fitting becomes formidable, while nonparametric modeling grapples with the curse of dimensionality. We propose a novel Penalized Deep Partially Linear Cox Model (Penalized DPLC), which incorporates the smoothly clipped absolute deviation (SCAD) penalty to select important texture features and employs a deep neural network to estimate the nonparametric component of the model. We prove the convergence and asymptotic properties of the estimator and compare it to other methods through extensive simulation studies, evaluating its performance in risk prediction and feature selection. The proposed method is applied to the NLST study dataset to uncover the effects of key clinical and imaging risk factors on patients' survival. Our findings provide valuable insights into the relationship between these factors and survival outcomes.

Kernel Cox partially linear regression: Building predictive models for cancer patients' survival.

Wide heterogeneity exists in cancer patients' survival, ranging from a few months to several decades. To accurately predict clinical outcomes, it is vital to build an accurate predictive model that relates the patients' molecular profiles with the patients' survival. With complex relationships between survival and high-dimensional molecular predictors, it is challenging to conduct nonparametric modeling and irrelevant predictors removing simultaneously. In this article, we build a kernel Cox proportional hazards semi-parametric model and propose a novel regularized garrotized kernel machine (RegGKM) method to fit the model. We use the kernel machine method to describe the complex relationship between survival and predictors, while automatically removing irrelevant parametric and nonparametric predictors through a LASSO penalty. An efficient high-dimensional algorithm is developed for the proposed method. Comparison with other competing methods in simulation shows that the proposed method always has better predictive accuracy. We apply this method to analyze a multiple myeloma dataset and predict the patients' death burden based on their gene expressions. Our results can help classify patients into groups with different death risks, facilitating treatment for better clinical outcomes.

The role of Matrix Metalloproteinase-2 and Galectin-3 as predictive biomarkers for all-cause mortality in patients undergoing transfemoral transcatheter aortic valve implantation.

BACKGROUND: Currently available risk scores fail to accurately predict morbidity and mortality in patients with severe symptomatic aortic stenosis who undergo transcatheter aortic valve implantation (TAVI). In this context, biomarkers like matrix metalloproteinase-2 (MMP-2) and Galectin-3 (Gal-3) may provide additional prognostic information. METHODS: Patients with severe aortic stenosis undergoing consecutive, elective, transfemoral TAVI were included. Baseline demographic data, functional status, echocardiographic findings, clinical outcomes and biomarker levels were collected and analysed. RESULTS: The study cohort consisted of 89 patients (age 80.4 ± 5.1 years, EuroScore II 7.1 ± 5.8%). During a median follow-up period of 526 d, 28 patients (31.4%) died. Among those who died, median baseline MMP-2 (alive: 221.6 [170.4; 263] pg/mL vs. deceased: 272.1 [225; 308.8] pg/mL, p < 0.001) and Gal-3 levels (alive: 19.1 [13.5; 24.6] pg/mL vs. deceased: 25 [17.6; 29.5] pg/mL, p = 0.006) were higher than in survivors. In ROC analysis, MMP-2 reached an acceptable level of discrimination to predict mortality (AUC 0.733, 95% CI [0.62; 0.83], p < 0.001), but the predictive value of Gal-3 was poor (AUC 0.677, 95% CI [0.56; 0.79], p = 0.002). Kaplan-Meier and Cox regression analyses showed that patients with MMP-2 and Gal-3 concentrations above the median at baseline had significantly impaired long-term survival (p = 0.004 and p = 0.02, respectively). CONCLUSIONS: In patients with severe aortic stenosis undergoing transfemoral TAVI, MMP-2 and to a lesser extent Gal-3, seem to have additive value in optimizing risk prediction and streamlining decision-making.

Admission blood tests predicting survival of SARS-CoV-2 infected patients: a practical implementation of graph convolution network in imbalance dataset.

BACKGROUND: Predicting an individual's risk of death from COVID-19 is essential for planning and optimising resources. However, since the real-world mortality rate is relatively low, particularly in places like Hong Kong, this makes building an accurate prediction model difficult due to the imbalanced nature of the dataset. This study introduces an innovative application of graph convolutional networks (GCNs) to predict COVID-19 patient survival using a highly imbalanced dataset. Unlike traditional models, GCNs leverage structural relationships within the data, enhancing predictive accuracy and robustness. By integrating demographic and laboratory data into a GCN framework, our approach addresses class imbalance and demonstrates significant improvements in prediction accuracy. METHODS: The cohort included all consecutive positive COVID-19 patients fulfilling study criteria admitted to 42 public hospitals in Hong Kong between January 23 and December 31, 2020 (n = 7,606). We proposed the population-based graph convolutional neural network (GCN) model which took blood test results, age and sex as inputs to predict the survival outcomes. Furthermore, we compared our proposed model to the Cox Proportional Hazard (CPH) model, conventional machine learning models, and oversampling machine learning models. Additionally, a subgroup analysis was performed on the test set in order to acquire a deeper understanding of the relationship between each patient node and its neighbours, revealing possible underlying causes of the inaccurate predictions. RESULTS: The GCN model was the top-performing model, with an AUC of 0.944, considerably outperforming all other models (p < 0.05), including the oversampled CPH model (0.708), linear regression (0.877), Linear Discriminant Analysis (0.860), K-nearest neighbours (0.834), Gaussian predictor (0.745) and support vector machine (0.847). With Kaplan-Meier estimates, the GCN model demonstrated good discriminability between low- and high-risk individuals (p < 0.0001). Based on subanalysis using the weighted-in score, although the GCN model was able to discriminate well between different predicted groups, the separation was inadequate between false negative (FN) and true negative (TN) groups. CONCLUSION: The GCN model considerably outperformed all other machine learning methods and baseline CPH models. Thus, when applied to this imbalanced COVID survival dataset, adopting a population graph representation may be an approach to achieving good prediction.

TES and SLC40A1 as Potential Biomarkers for Predicting Survival in T-Cell Acute Lymphoblastic Leukemia.

INTRODUCTION: Identifying patients with high-risk T-cell acute lymphoblastic leukemia (T-ALL) is crucial for personalized therapy; however, the lack of robust biomarkers hinders prognosis assessment. To address this issue, our study aimed to screen and validate genes whose expression may serve as predictive indicators of outcomes in T-ALL patients while also investigating the underlying molecular mechanisms. METHODS: Differentially expressed genes (DEGs) between T-ALL patients and healthy controls were identified by integrating data from three independent public datasets. Functional annotation of these DEGs and protein-protein interactions were also conducted. Further, we enrolled a prospective cohort of T-ALL patients (n = 20) at our center, conducting RNA-seq analysis on their bone marrow samples. Survival-based univariate Cox analysis was employed to identify gene expressions related to survival, and an intersection algorithm was sequentially applied. Furthermore, we validated the identified genes using cases from the Therapeutically Applicable Research to Generate Effective Treatments database, plotting Kaplan-Meier curves for secondary validation. RESULTS: Through the integration of survival-related genes with DEGs identified in T-ALL, our analysis revealed six T-ALL-specific genes, the expression levels of which were linked to prognostic value. Notably, the independent prognostic value of SLC40A1 and TES expression levels was confirmed in both an external cohort and a prospective cohort at our center. CONCLUSION: In summary, our preliminary study indicates that the expression levels of TES and SLC40A1 genes show promise as potential indicators for predicting survival outcomes in T-ALL patients.

Comparison of deep learning models to traditional Cox regression in predicting survival of colon cancer: Based on the SEER database.

BACKGROUND AND AIM: In this study, a deep learning algorithm was used to predict the survival rate of colon cancer (CC) patients, and compared its performance with traditional Cox regression. METHODS: In this population-based cohort study, we used the characteristics of patients diagnosed with CC between 2010 and 2015 from the Surveillance, Epidemiology and End Results (SEER) database. The population was randomized into a training set (n = 10 596, 70%) and a test set (n = 4536, 30%). Brier scores, area under the (AUC) receiver operating characteristic curve and calibration curves were used to compare the performance of the three most popular deep learning models, namely, artificial neural networks (ANN), deep neural networks (DNN), and long-short term memory (LSTM) neural networks with Cox proportional hazard (CPH) model. RESULTS: In the independent test set, the Brier values of ANN, DNN, LSTM and CPH were 0.155, 0.149, 0.148, and 0.170, respectively. The AUC values were 0.906 (95% confidence interval [CI] 0.897-0.916), 0.908 (95% CI 0.899-0.918), 0.910 (95% CI 0.901-0.919), and 0.793 (95% CI 0.769-0.816), respectively. Deep learning showed superior promising results than CPH in predicting CC specific survival. CONCLUSIONS: Deep learning showed potential advantages over traditional CPH models in terms of prognostic assessment and treatment recommendations. LSTM exhibited optimal predictive accuracy and has the ability to provide reliable information on individual survival and treatment recommendations for CC patients.

Analysis of the survival time of patients with heart failure with reduced ejection fraction: a Bayesian approach via a competing risk parametric model.

PURPOSE: Heart failure (HF) is a widespread ailment and is a primary contributor to hospital admissions. The focus of this study was to identify factors affecting the extended-term survival of patients with HF, anticipate patient outcomes through cause-of-death analysis, and identify risk elements for preventive measures. METHODS: A total of 435 HF patients were enrolled from the medical records of the Rajaie Cardiovascular Medical and Research Center, covering data collected between March and August 2018. After a five-year follow-up (July 2023), patient outcomes were assessed based on the cause of death. The survival analysis was performed with the AFT method with the Bayesian approach in the presence of competing risks. RESULTS: Based on the results of the best model for HF-related mortality, age [time ratio = 0.98, confidence interval 95%: 0.96-0.99] and ADHF [TR = 0.11, 95% (CI): 0.01-0.44] were associated with a lower survival time. Chest pain in HF-related mortality [TR = 0.41, 95% (CI): 0.10-0.96] and in non-HF-related mortality [TR = 0.38, 95% (CI): 0.12-0.86] was associated with a lower survival time. The next significant variable in HF-related mortality was hyperlipidemia (yes): [TR = 0.34, 95% (CI): 0.13-0.64], and in non-HF-related mortality hyperlipidemia (yes): [TR = 0.60, 95% (CI): 0.37-0.90]. CAD [TR = 0.65, 95% (CI): 0.38-0.98], CKD [TR = 0.52, 95% (CI): 0.28-0.87], and AF [TR = 0.53, 95% (CI): 0.32-0.81] were other variables that were directly related to the reduction in survival time of patients with non-HF-related mortality. CONCLUSION: The study identified distinct predictive factors for overall survival among patients with HF-related mortality or non-HF-related mortality. This differentiated approach based on the cause of death contributes to the estimation of patient survival time and provides valuable insights for clinical decision-making.

Mortality risk factors in patients receiving ECPR after cardiac arrest: Development and validation of a clinical prognostic prediction model.

BACKGROUND: Previous studies have shown an increasing trend of extracorporeal cardiopulmonary resuscitation (ECPR) use in patients with cardiac arrest (CA). Although ECPR have been found to reduce mortality in patients with CA compared with conventional cardiopulmonary resuscitation (CCPR), the mortality remains high. This study was designed to identify the potential mortality risk factors for ECPR patients for further optimization of patient management and treatment selection. METHODS: We conducted a prospective, multicentre study collecting 990 CA patients undergoing ECPR in 61 hospitals in China from January 2017 to May 2022 in CSECLS registry database. A clinical prediction model was developed using cox regression and validated with external data. RESULTS: The data of 351 patients meeting the inclusion criteria before October 2021 was used to develop a prediction model and that of 68 patients after October 2021 for validation. Of the 351 patients with CA treated with ECPR, 227 (64.8%) patients died before hospital discharge. Multivariate analysis suggested that a medical history of cerebrovascular diseases, pulseless electrical activity (PEA)/asystole and higher Lactate (Lac) were risk factors for mortality while aged 45-60, higher pH and intra-aortic balloon pump (IABP) during ECPR have protective effects. Internal validation by bootstrap resampling was subsequently used to evaluate the stability of the model, showing moderate discrimination, especially in the early stage following ECPR, with a C statistic of 0.70 and adequate calibration with GOF chi-square = 10.4 (p = 0.50) for the entire cohort. Fair discrimination with c statistic of 0.65 and good calibration (GOF chi-square = 6.1, p = 0.809) in the external validation cohort demonstrating the model's ability to predict in-hospital death across a wide range of probabilities. CONCLUSION: Risk factors have been identified among ECPR patients including a history of cerebrovascular diseases, higher Lac and presence of PEA or asystole. While factor such as age 45-60, higher pH and use of IABP have been found protective against in-hospital mortality. These factors can be used for risk prediction, thereby improving the management and treatment selection of patients for this resource-intensive therapy.

Comment on, "Survival prediction of glioblastoma patients-are we there yet? A systematic review of prognostic modeling for glioblastoma and its clinical potential".

The article "Survival Prediction of Glioblastoma Patients-Are We There Yet? A Systematic Review of Prognostic Modeling for Glioblastoma and Its Clinical Potential" by Tewarie et al. (2024) critically examines the current landscape of prognostic models for glioblastoma, highlighting both advancements and challenges in their clinical application. Through a systematic review adhering to PRISMA guidelines, the authors synthesize findings from diverse studies, shedding light on the variability in model performance and the obstacles to clinical implementation. Despite these contributions, the review faces limitations due to the heterogeneity of the studies included, which complicates definitive conclusions. The authors emphasize the need for external validation and standardization, though further exploration of the persistence of these challenges and the biases in machine learning models is warranted. Future research should focus on standardizing protocols and integrating ethical considerations to enhance the clinical utility of these models, moving the field closer to practical application.

Multiomics-Based Feature Extraction and Selection for the Prediction of Lung Cancer Survival.

Lung cancer is a global health challenge, hindered by delayed diagnosis and the disease's complex molecular landscape. Accurate patient survival prediction is critical, motivating the exploration of various -omics datasets using machine learning methods. Leveraging multi-omics data, this study seeks to enhance the accuracy of survival prediction by proposing new feature extraction techniques combined with unbiased feature selection. Two lung adenocarcinoma multi-omics datasets, originating from the TCGA and CPTAC-3 projects, were employed for this purpose, emphasizing gene expression, methylation, and mutations as the most relevant data sources that provide features for the survival prediction models. Additionally, gene set aggregation was shown to be the most effective feature extraction method for mutation and copy number variation data. Using the TCGA dataset, we identified 32 molecular features that allowed the construction of a 2-year survival prediction model with an AUC of 0.839. The selected features were additionally tested on an independent CPTAC-3 dataset, achieving an AUC of 0.815 in nested cross-validation, which confirmed the robustness of the identified features.

Prediction of survival of persons with advanced dementia using the advanced dementia prognostic tool: A 2-year prospective study.

BACKGROUND: In this prospective study, we evaluated the usefulness of the advanced dementia prognostic tool (ADEPT) for estimating the 2-year survival of persons with advanced dementia (AD) in China. METHODS: The study predicted the 2-year mortality of 115 persons with AD using the ADEPT score. RESULTS: In total, 115 persons with AD were included in the study. Of these persons, 48 died. The mean ADEPT score was 13.0. The AUROC for the prediction of the 2-year mortality rate using the ADEPT score was 0.62. The optimal threshold of the ADEPT score was 11.2, which had an AUROC of 0.63, specificity of 41.8, and sensitivity of 83.3. CONCLUSIONS: The ADEPT score based on a threshold of 11.2 may serve as a prognostic tool to determine the 2-year survival rate of persons with AD in Chongqing, China. However, further studies are needed to explore the nature of this relationship.

Cancer survival prediction by learning comprehensive deep feature representation for multiple types of genetic data.

BACKGROUND: Cancer is one of the leading death causes around the world. Accurate prediction of its survival time is significant, which can help clinicians make appropriate therapeutic schemes. Cancer data can be characterized by varied molecular features, clinical behaviors and morphological appearances. However, the cancer heterogeneity problem usually makes patient samples with different risks (i.e., short and long survival time) inseparable, thereby causing unsatisfactory prediction results. Clinical studies have shown that genetic data tends to contain more molecular biomarkers associated with cancer, and hence integrating multi-type genetic data may be a feasible way to deal with cancer heterogeneity. Although multi-type gene data have been used in the existing work, how to learn more effective features for cancer survival prediction has not been well studied. RESULTS: To this end, we propose a deep learning approach to reduce the negative impact of cancer heterogeneity and improve the cancer survival prediction effect. It represents each type of genetic data as the shared and specific features, which can capture the consensus and complementary information among all types of data. We collect mRNA expression, DNA methylation and microRNA expression data for four cancers to conduct experiments. CONCLUSIONS: Experimental results demonstrate that our approach substantially outperforms established integrative methods and is effective for cancer survival prediction. AVAILABILITY AND IMPLEMENTATION: https://github.com/githyr/ComprehensiveSurvival .

Deep learning model integrating positron emission tomography and clinical data for prognosis prediction in non-small cell lung cancer patients.

BACKGROUND: Lung cancer is the leading cause of cancer-related deaths worldwide. The majority of lung cancers are non-small cell lung cancer (NSCLC), accounting for approximately 85% of all lung cancer types. The Cox proportional hazards model (CPH), which is the standard method for survival analysis, has several limitations. The purpose of our study was to improve survival prediction in patients with NSCLC by incorporating prognostic information from F-18 fluorodeoxyglucose positron emission tomography (FDG PET) images into a traditional survival prediction model using clinical data. RESULTS: The multimodal deep learning model showed the best performance, with a C-index and mean absolute error of 0.756 and 399 days under a five-fold cross-validation, respectively, followed by ResNet3D for PET (0.749 and 405 days) and CPH for clinical data (0.747 and 583 days). CONCLUSION: The proposed deep learning-based integrative model combining the two modalities improved the survival prediction in patients with NSCLC.

Development and validation of a deep learning survival model for cervical adenocarcinoma patients.

BACKGROUND: The aim was to develop a personalized survival prediction deep learning model for cervical adenocarcinoma patients and process personalized survival prediction. METHODS: A total of 2501 cervical adenocarcinoma patients from the surveillance, epidemiology and end results database and 220 patients from Qilu hospital were enrolled in this study. We created our deep learning (DL) model to manipulate the data and evaluated its performance against four other competitive models. We tried to demonstrate a new grouping system oriented by survival outcomes and process personalized survival prediction by using our DL model. RESULTS: The DL model reached 0.878 c-index and 0.09 Brier score in the test set, which was better than the other four models. In the external test set, our model achieved a 0.80 c-index and 0.13 Brier score. Thus, we developed prognosis-oriented risk grouping for patients according to risk scores computed by our DL model. Notable differences among groupings were observed. In addition, a personalized survival prediction system based on our risk-scoring grouping was developed. CONCLUSIONS: We developed a deep neural network model for cervical adenocarcinoma patients. The performance of this model proved to be superior to other models. The results of external validation supported the possibility that the model can be used in clinical work. Finally, our survival grouping and personalized prediction system provided more accurate prognostic information for patients than traditional FIGO stages.

New insights into glioma frequency maps: From genetic and transcriptomic correlate to survival prediction.

Increasing evidence indicates that glioma topographic location is linked to the cellular origin, molecular alterations and genetic profile. This research aims to (a) reveal the underlying mechanisms of tumor location predilection in glioblastoma multiforme (GBM) and lower-grade glioma (LGG) and (b) leverage glioma location features to predict prognosis. MRI images from 396 GBM and 190 LGG (115 astrocytoma and 75 oligodendroglioma) patients were standardized to construct frequency maps and analyzed by voxel-based lesion-symptom mapping. We then investigated the spatial correlation between glioma distribution with gene expression in healthy brains. We also evaluated transcriptomic differences in tumor tissue from predilection and nonpredilection sites. Furthermore, we quantitively characterized tumor anatomical localization and explored whether it was significantly related to overall survival. Finally, we employed a support vector machine to build a survival prediction model for GBM patients. GBMs exhibited a distinct location predilection from LGGs. GBMs were nearer to the subventricular zone and more likely to be localized to regions enriched with synaptic signaling, whereas astrocytoma and oligodendroglioma tended to occur in areas associated with the immune response. Synapse, neurotransmitters and calcium ion channel-related genes were all activated in GBM tissues coming from predilection regions. Furthermore, we characterized tumor location features in terms of a series of tumor-to-predilection distance metrics, which were able to predict GBM 1-year survival status with an accuracy of 0.71. These findings provide new perspectives on our understanding of tumor anatomic localization. The spatial features of glioma are of great value in individual therapy and prognosis prediction.

Survival rate prediction of nasopharyngeal carcinoma patients based on MRI and gene expression using a deep neural network.

To achieve a better treatment regimen and follow-up assessment design for intensity-modulated radiotherapy (IMRT)-treated nasopharyngeal carcinoma (NPC) patients, an accurate progression-free survival (PFS) time prediction algorithm is needed. We propose developing a PFS prediction model of NPC patients after IMRT treatment using a deep learning method and comparing that with the traditional texture analysis method. One hundred and fifty-one NPC patients were included in this retrospective study. T1-weighted, proton density and dynamic contrast-enhanced magnetic resonance (MR) images were acquired. The expression level of five genes (HIF-1α, EGFR, PTEN, Ki-67, and VEGF) and infection of Epstein-Barr (EB) virus were tested. A residual network was trained to predict PFS from MR images. The output as well as patient characteristics were combined using a linear regression model to provide a final PFS prediction. The prediction accuracy was compared with that of the traditional texture analysis method. A regression model combining the deep learning output with HIF-1α expression and Epstein-Barr infection provides the best PFS prediction accuracy (Spearman correlation R2  = 0.53; Harrell's C-index = 0.82; receiver operative curve [ROC] analysis area under the curve [AUC] = 0.88; log-rank test hazard ratio [HR] = 8.45), higher than a regression model combining texture analysis with HIF-1α expression (Spearman correlation R2  = 0.14; Harrell's C-index =0.68; ROC analysis AUC = 0.76; log-rank test HR = 2.85). The deep learning method does not require a manually drawn tumor region of interest. MR image processing using deep learning combined with patient characteristics can provide accurate PFS prediction for nasopharyngeal carcinoma patients and does not rely on specific kernels or tumor regions of interest, which is needed for the texture analysis method.

Performance evaluation of transcriptomics data normalization for survival risk prediction.

One pivotal feature of transcriptomics data is the unwanted variations caused by disparate experimental handling, known as handling effects. Various data normalization methods were developed to alleviate the adverse impact of handling effects in the setting of differential expression analysis. However, little research has been done to evaluate their performance in the setting of survival outcome prediction, an important analysis goal for transcriptomics data in biomedical research. Leveraging a unique pair of datasets for the same set of tumor samples-one with handling effects and the other without, we developed a benchmarking tool for conducting such an evaluation in microRNA microarrays. We applied this tool to evaluate the performance of three popular normalization methods-quantile normalization, median normalization and variance stabilizing normalization-in survival prediction using various approaches for model building and designs for sample assignment. We showed that handling effects can have a strong impact on survival prediction and that quantile normalization, a most popular method in current practice, tends to underperform median normalization and variance stabilizing normalization. We demonstrated with a small example the reason for quantile normalization's poor performance in this setting. Our finding highlights the importance of putting normalization evaluation in the context of the downstream analysis setting and the potential of improving the development of survival predictors by applying median normalization. We make available our benchmarking tool for performing such evaluation on additional normalization methods in connection with prediction modeling approaches.