Training artificial neural networks directly on the concordance index for censored data using genetic algorithms

Artificial Intelligence in Coronary Artery Calcium Scoring

Cardiovascular disease (CVD), particularly coronary heart disease (CHD), is the leading cause of death in the US, with a high economic impact. Coronary artery calcium (CAC) is a known marker for CHD and a useful tool for estimating the risk of atherosclerotic cardiovascular disease (ASCVD). Although CACS is recommended for informing the decision to initiate statin therapy, the current standard requires a dedicated CT protocol, which is time-intensive and contributes to radiation exposure. Non-dedicated CT protocols can be taken advantage of to visualize calcium and reduce overall cost and radiation exposure; however, they mainly provide visual estimates of coronary calcium and have disadvantages such as motion artifacts. Artificial intelligence is a growing field involving software that independently performs human-level tasks, and is well suited for improving CACS efficiency and repurposing non-dedicated CT for calcium scoring. We present a review of the current studies on automated CACS across various CT protocols and discuss consideration points in clinical application and some barriers to implementation.

Deep Learning Can Predict Bevacizumab Therapeutic Effect and Microsatellite Instability Directly from Histology in Epithelial Ovarian Cancer

Epithelial ovarian cancer (EOC) remains a significant cause of mortality among gynecologic cancers, with the majority of cases being diagnosed at an advanced stage. Before targeted therapies were available, EOC treatment relied largely on debulking surgery and platinum-based chemotherapy. Vascular endothelial growth factors have been identified as inducing tumor angiogenesis. According to several clinical trials, anti-vascular endothelial growth factor-targeted therapy with bevacizumab was effective in all phases of EOC treatment. However, there are currently no biomarkers accessible for regular therapeutic use despite the importance of patient selection. Microsatellite instability (MSI), caused by a deficiency of the DNA mismatch repair system, is a molecular abnormality observed in EOC associated with Lynch syndrome. Recent evidence suggests that angiogenesis and MSI are interconnected. Developing predictive biomarkers, which enable the selection of patients who might benefit from bevacizumab-targeted therapy or immunotherapy, is critical for realizing personalized precision medicine. In this study, we developed 2 improved deep learning methods that eliminate the need for laborious detailed image-wise annotations by pathologists and compared them with 3 state-of-the-art methods to not only predict the efficacy of bevacizumab in patients with EOC using mismatch repair protein immunostained tissue microarrays but also predict MSI status directly from histopathologic images. In prediction of therapeutic outcomes, the 2 proposed methods achieved excellent performance by obtaining the highest mean sensitivity and specificity score using MSH2 or MSH6 markers and outperformed 3 state-of-the-art deep learning methods. Moreover, both statistical analysis results, using Cox proportional hazards model analysis and Kaplan-Meier progression-free survival analysis, confirm that the 2 proposed methods successfully differentiate patients with positive therapeutic effects and lower cancer recurrence rates from patients experiencing disease progression after treatment (P < .01). In prediction of MSI status directly from histopathology images, our proposed method also achieved a decent performance in terms of mean sensitivity and specificity score even for imbalanced data sets for both internal validation using tissue microarrays from the local hospital and external validation using whole section slides from The Cancer Genome Atlas archive.

Development and validation of a novel risk score to predict 5-year mortality in patients with acute myocardial infarction in China: a retrospective study

BACKGROUND

The disease burden from ischaemic heart disease remains heavy in the Chinese population. Traditional risk scores for estimating long-term mortality in patients with acute myocardial infarction (AMI) have been developed without sufficiently considering advances in interventional procedures and medication. The goal of this study was to develop a risk score comprising clinical parameters and intervention advances at hospital admission to assess 5-year mortality in AMI patients in a Chinese population.

METHODS

We performed a retrospective observational study on 2,722 AMI patients between January 2013 and December 2017. Of these patients, 1,471 patients from Changsha city, Hunan Province, China were assigned to the development cohort, and 1,251 patients from Xiangtan city, Hunan Province, China, were assigned to the validation cohort. Forty-five candidate variables assessed at admission were screened using least absolute shrinkage and selection operator, stepwise backward regression, and Cox regression methods to construct the C2ABS2-GLPK score, which was graded and stratified using a nomogram and X-tile. The score was internally and externally validated. The C-statistic and Hosmer-Lemeshow test were used to assess discrimination and calibration, respectively.

RESULTS

From the 45 candidate variables obtained at admission, 10 potential predictors, namely, including Creatinine, experience of Cardiac arrest, Age, N-terminal Pro-Brain Natriuretic Peptide, a history of Stroke, Statins therapy, fasting blood Glucose, Left ventricular end-diastolic diameter, Percutaneous coronary intervention and Killip classification were identified as having a close association with 5-year mortality in patients with AMI and collectively termed the C2ABS2-GLPK score. The score had good discrimination (C-statistic = 0.811, 95% confidence intervals (CI) [0.786-0.836]) and calibration (calibration slope = 0.988) in the development cohort. In the external validation cohort, the score performed well in both discrimination (C-statistic = 0.787, 95% CI [0.756-0.818]) and calibration (calibration slope = 0.976). The patients were stratified into low- (≤148), medium- (149 to 218) and high-risk (≥219) categories according to the C2ABS2-GLPK score. The predictive performance of the score was also validated in all subpopulations of both cohorts.

CONCLUSION

The C2ABS2-GLPK score is a Chinese population-based risk assessment tool to predict 5-year mortality in AMI patients based on 10 variables that are routinely assessed at admission. This score can assist physicians in stratifying high-risk patients and optimizing emergency medical interventions to improve long-term survival in patients with AMI.

SurvNet: A Novel Deep Neural Network for Lung Cancer Survival Analysis With Missing Values

Survival analysis is important for guiding further treatment and improving lung cancer prognosis. It is a challenging task because of the poor distinguishability of features and the missing values in practice. A novel multi-task based neural network, SurvNet, is proposed in this paper. The proposed SurvNet model is trained in a multi-task learning framework to jointly learn across three related tasks: input reconstruction, survival classification, and Cox regression. It uses an input reconstruction mechanism cooperating with incomplete-aware reconstruction loss for latent feature learning of incomplete data with missing values. Besides, the SurvNet model introduces a context gating mechanism to bridge the gap between survival classification and Cox regression. A new real-world dataset of 1,137 patients with IB-IIA stage non-small cell lung cancer is collected to evaluate the performance of the SurvNet model. The proposed SurvNet achieves a higher concordance index than the traditional Cox model and Cox-Net. The difference between high-risk and low-risk groups obtained by SurvNet is more significant than that of high-risk and low-risk groups obtained by the other models. Moreover, the SurvNet outperforms the other models even though the input data is randomly cropped and it achieves better generalization performance on the Surveillance, Epidemiology, and End Results Program (SEER) dataset.

A 21‑gene Support Vector Machine classifier and a 10‑gene risk score system constructed for patients with gastric cancer

Gastric cancer (GC) ranks fifth in terms of incidence and third in terms of tumor mortality worldwide. The present study was designed to construct a Support Vector Machine (SVM) classifier and risk score system for GC. The GSE62254 (training set) and GSE26253 (validation set 2) datasets were downloaded from the Gene Expression Omnibus database. Furthermore, the gene expression profile of GC (validation set 1) was obtained from The Cancer Genome Atlas database. Differentially expressed genes (DEGs) between recurrent and non‑recurrent samples were determined using the limma package. The feature genes were selected using the Caret package, and an SVM classifier was built using the e1071 package. Using the penalized package, the optimal predictive genes for constructing a risk score system were screened. Finally, stratification analysis of clinical factors and pathway enrichment analysis were performed using Gene Set Enrichment Analysis. A total of 239 DEGs were identified in GSE62254, among which 114 DEGs were significantly associated with both recurrence‑free survival and overall survival. Subsequently, 21 feature genes were screened from the 114 DEGs, and an SVM classifier was built. A risk score system for survival prediction was constructed, following the selection of 10 optimal genes, including A‑kinase anchoring protein 12, angiopoietin‑like protein 1, cysteine‑rich sequence 1, myeloid/lymphoid or mixed‑lineage leukemia, translocated to chromosome 11, neuron navigator 3, neurobeachin, nephroblastoma overexpressed, pleiotrophin, tumor suppressor candidate 3 and zinc finger and SCAN domain containing 18. The stratification analysis revealed that pathological stage was an independent prognostic clinical factor in the high‑risk group. Additionally, eight significant pathways were associated with the 10‑gene signature. The SVM classifier and risk score system may be applied for classifying and predicting the prognosis of patients with GC, respectively.

Non-linear transformations of age at diagnosis, tumor size, and number of positive lymph nodes in prediction of clinical outcome in breast cancer

BACKGROUND

Prognostic factors in breast cancer are often measured on a continuous scale, but categorized for clinical decision-making. The primary aim of this study was to evaluate if accounting for continuous non-linear effects of the three factors age at diagnosis, tumor size, and number of positive lymph nodes improves prognostication. These factors will most likely be included in the management of breast cancer patients also in the future, after an expected implementation of gene expression profiling for adjuvant treatment decision-making.

METHODS

Four thousand four hundred forty seven and 1132 women with primary breast cancer constituted the derivation and validation set, respectively. Potential non-linear effects on the log hazard of distant recurrences of the three factors were evaluated during 10 years of follow-up. Cox-models of successively increasing complexity: dichotomized predictors, predictors categorized into three or four groups, and predictors transformed using fractional polynomials (FPs) or restricted cubic splines (RCS), were used. Predictive performance was evaluated by Harrell's C-index.

RESULTS

Using FP-transformations, non-linear effects were detected for tumor size and number of positive lymph nodes in univariable analyses. For age, non-linear transformations did, however, not improve the model fit significantly compared to the linear identity transformation. As expected, the C-index increased with increasing model complexity for multivariable models including the three factors. By allowing more than one cut-point per factor, the C-index increased from 0.628 to 0.674. The additional gain, as measured by the C-index, when using FP- or RCS-transformations was modest (0.695 and 0.696, respectively). The corresponding C-indices for these four models in the validation set, based on the same transformations and parameter estimates from the derivation set, were 0.675, 0.700, 0.706, and 0.701.

CONCLUSIONS

Categorization of each factor into three to four groups was found to improve prognostication compared to dichotomization. The additional gain by allowing continuous non-linear effects modeled by FPs or RCS was modest. However, the continuous nature of these transformations has the advantage of making it possible to form risk groups of any size.

Feature selection through validation and un-censoring of endovascular repair survival data for predicting the risk of re-intervention

Background

Feature selection (FS) process is essential in the medical area as it reduces the effort and time needed for physicians to measure unnecessary features. Choosing useful variables is a difficult task with the presence of censoring which is the unique characteristic in survival analysis. Most survival FS methods depend on Cox’s proportional hazard model; however, machine learning techniques (MLT) are preferred but not commonly used due to censoring. Techniques that have been proposed to adopt MLT to perform FS with survival data cannot be used with the high level of censoring. The researcher’s previous publications proposed a technique to deal with the high level of censoring. It also used existing FS techniques to reduce dataset dimension. However, in this paper a new FS technique was proposed and combined with feature transformation and the proposed uncensoring approaches to select a reduced set of features and produce a stable predictive model.

Methods

In this paper, a FS technique based on artificial neural network (ANN) MLT is proposed to deal with highly censored Endovascular Aortic Repair (EVAR). Survival data EVAR datasets were collected during 2004 to 2010 from two vascular centers in order to produce a final stable model. They contain almost 91% of censored patients. The proposed approach used a wrapper FS method with ANN to select a reduced subset of features that predict the risk of EVAR re-intervention after 5 years to patients from two different centers located in the United Kingdom, to allow it to be potentially applied to cross-centers predictions. The proposed model is compared with the two popular FS techniques; Akaike and Bayesian information criteria (AIC, BIC) that are used with Cox’s model.

Results

The final model outperforms other methods in distinguishing the high and low risk groups; as they both have concordance index and estimated AUC better than the Cox’s model based on AIC, BIC, Lasso, and SCAD approaches. These models have p-values lower than 0.05, meaning that patients with different risk groups can be separated significantly and those who would need re-intervention can be correctly predicted.

Conclusion

The proposed approach will save time and effort made by physicians to collect unnecessary variables. The final reduced model was able to predict the long-term risk of aortic complications after EVAR. This predictive model can help clinicians decide patients’ future observation plan.

Electronic supplementary material

The online version of this article (doi:10.1186/s12911-017-0508-3) contains supplementary material, which is available to authorized users.

A framework for parameter estimation and model selection in kernel deep stacking networks

BACKGROUND AND OBJECTIVES

Kernel deep stacking networks (KDSNs) are a novel method for supervised learning in biomedical research. Belonging to the class of deep learning techniques, KDSNs are based on artificial neural network architectures that involve multiple nonlinear transformations of the input data. Unlike traditional artificial neural networks, KDSNs do not rely on backpropagation algorithms but on an efficient fitting procedure that is based on a series of kernel ridge regression models with closed-form solutions. Although being computationally advantageous, KDSN modeling remains a challenging task, as it requires the specification of a large number of tuning parameters.

METHODS AND MATERIAL

We propose a new data-driven framework for parameter estimation, hyperparameter tuning, and model selection in KDSNs. The proposed methodology is based on a combination of model-based optimization and hill climbing approaches that do not require the pre-specification of any of the KDSN tuning parameters. We demonstrate the performance of KDSNs by analyzing three medical data sets on hospital readmission of diabetes patients, coronary artery disease, and hospital costs.

RESULTS

Our numerical studies show that the run-time of the proposed KDSN methodology is significantly shorter than the respective run-time of grid search strategies for hyperparameter tuning. They also show that KDSN modeling is competitive in terms of prediction accuracy with other state-of-the-art techniques for statistical learning.

CONCLUSIONS

KDSNs are a computationally efficient approximation of backpropagation-based artificial neural network techniques. Application of the proposed methodology results in a fast tuning procedure that generates KDSN fits having a similar prediction accuracy as other techniques in the field of deep learning.

Finding Risk Groups by Optimizing Artificial Neural Networks on the Area under the Survival Curve Using Genetic Algorithms

We investigate a new method to place patients into risk groups in censored survival data. Properties such as median survival time, and end survival rate, are implicitly improved by optimizing the area under the survival curve. Artificial neural networks (ANN) are trained to either maximize or minimize this area using a genetic algorithm, and combined into an ensemble to predict one of low, intermediate, or high risk groups. Estimated patient risk can influence treatment choices, and is important for study stratification. A common approach is to sort the patients according to a prognostic index and then group them along the quartile limits. The Cox proportional hazards model (Cox) is one example of this approach. Another method of doing risk grouping is recursive partitioning (Rpart), which constructs a decision tree where each branch point maximizes the statistical separation between the groups. ANN, Cox, and Rpart are compared on five publicly available data sets with varying properties. Cross-validation, as well as separate test sets, are used to validate the models. Results on the test sets show comparable performance, except for the smallest data set where Rpart’s predicted risk groups turn out to be inverted, an example of crossing survival curves. Cross-validation shows that all three models exhibit crossing of some survival curves on this small data set but that the ANN model manages the best separation of groups in terms of median survival time before such crossings. The conclusion is that optimizing the area under the survival curve is a viable approach to identify risk groups. Training ANNs to optimize this area combines two key strengths from both prognostic indices and Rpart. First, a desired minimum group size can be specified, as for a prognostic index. Second, the ability to utilize non-linear effects among the covariates, which Rpart is also able to do.

Raman spectroscopy of blood serum for Alzheimer's disease diagnostics: specificity relative to other types of dementia

The key moment for efficiently and accurately diagnosing dementia occurs during the early stages. This is particularly true for Alzheimer's disease (AD). In this proof-of-concept study, we applied near infrared (NIR) Raman microspectroscopy of blood serum together with advanced multivariate statistics for the selective identification of AD. We analyzed data from 20 AD patients, 18 patients with other neurodegenerative dementias (OD) and 10 healthy control (HC) subjects. NIR Raman microspectroscopy differentiated patients with more than 95% sensitivity and specificity. We demonstrated the high discriminative power of artificial neural network (ANN) classification models, thus revealing the high potential of this developed methodology for the differential diagnosis of AD. Raman spectroscopic, blood-based tests may aid clinical assessments for the effective and accurate differential diagnosis of AD, decrease the labor, time and cost of diagnosis, and be useful for screening patient populations for AD development and progression. Multivariate data analysis of blood serum Raman spectra allows for the differentiation between patients with Alzheimer's disease, other types of dementia and healthy individuals.

Bayesian neural network approach for determining the risk of re-intervention after endovascular aortic aneurysm repair

This article proposes a Bayesian neural network approach to determine the risk of re-intervention after endovascular aortic aneurysm repair surgery. The target of proposed technique is to determine which patients have high chance to re-intervention (high-risk patients) and which are not (low-risk patients) after 5 years of the surgery. Two censored datasets relating to the clinical conditions of aortic aneurysms have been collected from two different vascular centers in the United Kingdom. A Bayesian network was first employed to solve the censoring issue in the datasets. Then, a back propagation neural network model was built using the uncensored data of the first center to predict re-intervention on the second center and classify the patients into high-risk and low-risk groups. Kaplan-Meier curves were plotted for each group of patients separately to show whether there is a significant difference between the two risk groups. Finally, the logrank test was applied to determine whether the neural network model was capable of predicting and distinguishing between the two risk groups. The results show that the Bayesian network used for uncensoring the data has improved the performance of the neural networks that were built for the two centers separately. More importantly, the neural network that was trained with uncensored data of the first center was able to predict and discriminate between groups of low risk and high risk of re-intervention after 5 years of endovascular aortic aneurysm surgery at center 2 (p = 0.0037 in the logrank test).

Analysis of regional bone scan index measurements for the survival of patients with prostate cancer

Background

A bone scan is a common method for monitoring bone metastases in patients with advanced prostate cancer. The Bone Scan Index (BSI) measures the tumor burden on the skeleton, expressed as a percentage of the total skeletal mass. Previous studies have shown that BSI is associated with survival of prostate cancer patients. The objective in this study was to investigate to what extent regional BSI measurements, as obtained by an automated method, can improve the survival analysis for advanced prostate cancer.

Methods

The automated method for analyzing bone scan images computed BSI values for twelve skeletal regions, in a study population consisting of 1013 patients diagnosed with prostate cancer. In the survival analysis we used the standard Cox proportional hazards model and a more advanced non-linear method based on artificial neural networks. The concordance index (C-index) was used to measure the performance of the models.

Results

A Cox model with age and total BSI obtained a C-index of 70.4%. The best Cox model with regional measurements from Costae, Pelvis, Scapula and the Spine, together with age, got a similar C-index (70.5%). The overall best single skeletal localisation, as measured by the C-index, was Costae. The non-linear model performed equally well as the Cox model, ruling out any significant non-linear interactions among the regional BSI measurements.

Conclusion

The present study showed that the localisation of bone metastases obtained from the bone scans in prostate cancer patients does not improve the performance of the survival models compared to models using the total BSI. However a ranking procedure indicated that some regions are more important than others.