Bayesian Neural Network Classifier Research Articles

Training with Small Medical Data: Robust Bayesian Neural Networks for Colon Cancer Overall Survival Prediction.

Fast and accurate cancer prognosis stratification models are essential for treatment designs. Large labeled patient data can power advanced deep learning models to obtain precise predictions. However, since fully labeled patient data are hard to acquire in practical scenarios, deep models are prone to make non-robust predictions biased toward data partition and model hyper-parameter selection. Given a small training set, we applied the systems biology feature selector in our previous study to avoid over-fitting and select 18 prognostic biomarkers. Combined with three other clinical features, we trained Bayesian binary classifiers to predict the 5-year overall survival (OS) of colon cancer patients in this study. Results showed that Bayesian models could provide better and more robust predictions compared to their non-Bayesian counterparts. Specifically, in terms of the area under the receiver operating characteristic curve (AUC), macro F1-score (maF1), and concordance index (CI), we found that the Bayesian bimodal neural network (late fusion) classifier (B-Bimodal) achieved the best results (AUC: 0.8083 ± 0.0736; maF1: 0.7300 ± 0.0659; CI: 0.7238 ± 0.0440). The single modal Bayesian neural network classifier (B-Concat) fed with concatenated patient data (early fusion) achieved slightly worse but more robust performance in terms of AUC and CI (AUC: 0.7105 ± 0.0692; maF1: 0.7156 ± 0.0690; CI: 0.6627 ± 0.0558). Such robustness is essential to training learning models with small medical data.

Hierarchical learning with backtracking algorithm based on the Visual Confusion Label Tree for large-scale image classification

In this paper, a hierarchical learning algorithm based on the Bayesian Neural Network classifier with backtracking is proposed to support large-scale image classification, where a Visual Confusion Label Tree is established for constructing a hierarchical structure for large numbers of categories in image datasets and determining the hierarchical learning tasks automatically. Specifically, the Visual Confusion Label Tree is established based on outputs of convolution neural network models. One parent node on the Visual Confusion Label Tree contains a set of sibling coarse-grained categories, and child nodes have several sets of fine-grained categories which are partitions of categories on the parent node. The proposed Hierarchical Bayesian Neural Network with backtracking algorithm can benefit from the hierarchical structure of the Visual Confusion Label Tree. Focusing on those confusion subsets instead of the entire set of categories makes the classification ability of the tree classifier stronger. The backtracking algorithm can utilize the uncertainty information captured from the Bayesian Neural Network to make a second classification to re-correct samples that were classified incorrectly in the previous classification process. Experiments on four large-scale datasets show that our tree classifier obtains a significant improvement over the state-of-the-art tree classifier, which have demonstrated the discriminative hierarchical structure of our Visual Confusion Label Tree and the effectiveness of our Hierarchical Bayesian Neural Network with backtracking algorithm.

Bayesian Neural Networks for Identification and Classification of Radio Frequency Transmitters Using Power Amplifiers’ Nonlinearity Signatures

The edge devices in an emerging Internet-of-Things (IoT) environment require comprehensive security measures that are within the power budget for ubiquitous computing. In this paper, a transmitter identification scheme consisting of a lightweight Bayesian neural network (BNN)-based classifier using raw time-domain data is presented. Evaluation is performed with data obtained in schematic-level simulation of high-efficiency CMOS power amplifier designs using a 65 nm process design kit (PDK). The Bayesian neural networks achieve 89.5% accuracy on the task of classifying six transmitters. Moreover, the BNN classifier is implemented on field-programmable gate array (FPGA) with parallel pseudo-Gaussian random number generators to achieve a throughput of more than 340,000 classifications per second, with average energy consumption for each classification task of 0.548 μJ. This low-power system enables comprehensive security for energy-constrained IoT devices and sensors.

The combination of four molecular markers improves thyroid cancer cytologic diagnosis and patient management.

BackgroundPapillary thyroid cancer is the most common endocrine malignancy. The most sensitive and specific diagnostic tool for thyroid nodule diagnosis is fine-needle aspiration (FNA) biopsy with cytological evaluation. Nevertheless, FNA biopsy is not always decisive leading to “indeterminate” or “suspicious” diagnoses in 10 %–30 % of cases. BRAF V600E detection is currently used as molecular test to improve the diagnosis of thyroid nodules, yet it lacks sensitivity. The aim of the present study was to identify novel molecular markers/computational models to improve the discrimination between benign and malignant thyroid lesions.MethodsWe collected 118 pre-operative thyroid FNA samples. All 118 FNA samples were characterized for the presence of the BRAF V600E mutation (exon15) by pyrosequencing and further assessed for mRNA expression of four genes (KIT, TC1, miR-222, miR-146b) by quantitative polymerase chain reaction. Computational models (Bayesian Neural Network Classifier, discriminant analysis) were built, and their ability to discriminate benign and malignant tumors were tested. Receiver operating characteristic (ROC) analysis was performed and principal component analysis was used for visualization purposes.ResultsIn total, 36/70 malignant samples carried the V600E mutation, while all 48 benign samples were wild type for BRAF exon15. The Bayesian neural network (BNN) and discriminant analysis, including the mRNA expression of the four genes (KIT, TC1, miR-222, miR-146b) showed a very strong predictive value (94.12 % and 92.16 %, respectively) in discriminating malignant from benign patients. The discriminant analysis showed a correct classification of 100 % of the samples in the malignant group, and 95 % by BNN. KIT and miR-146b showed the highest diagnostic accuracy of the ROC curve, with area under the curve values of 0.973 for KIT and 0.931 for miR-146b.ConclusionsThe four genes model proposed in this study proved to be highly discriminative of the malignant status compared with BRAF assessment alone. Its implementation in clinical practice can help in identifying malignant/benign nodules that would otherwise remain suspicious.Electronic supplementary materialThe online version of this article (doi:10.1186/s12885-015-1917-2) contains supplementary material, which is available to authorized users.

COMPARISON OF SUPPORT VECTOR MACHINES AND BAYESIAN NEURAL NETWORKS PERFORMANCE FOR BREAST TISSUES USING GEOSTATISTICAL FUNCTIONS IN MAMMOGRAPHIC IMAGES

Female breast cancer is a major cause of deaths in occidental countries. Computer-aided Detection (CAD) systems can aid radiologists to increase diagnostic accuracy. In this work, we present a comparison between two classifiers applied to the separation of normal and abnormal breast tissues from mammograms. The purpose of the comparison is to select the best prediction technique to be part of a CAD system. Each region of interest is classified through a Support Vector Machine (SVM) and a Bayesian Neural Network (BNN) as normal or abnormal region. SVM is a machine-learning method, based on the principle of structural risk minimization, which shows good performance when applied to data outside the training set. A Bayesian Neural Network is a classifier that joins traditional neural networks theory and Bayesian inference. We use a set of measures obtained by the application of the semivariogram, semimadogram, covariogram, and correlogram functions to the characterization of breast tissue as normal or abnormal. The results show that SVM presents best performance for the classification of breast tissues in mammographic images. The tests indicate that SVM has more generalization power than the BNN classifier. BNN has a sensibility of 76.19% and a specificity of 79.31%, while SVM presents a sensibility of 74.07% and a specificity of 98.77%. The accuracy rate for tests is 78.70% and 92.59% for BNN and SVM, respectively.

WE-C-332-05: Preliminary Robustness Study of a Breast Ultrasound Computer-Aided Diagnosis System Across Different Patient Populations

Purpose: To determine the robustness of a breast ultrasound computer‐aided diagnosis system (CADx) when it is used across different patient populations. Method and Materials: A sonographic database consisting of 433 lesions (127 malignant, 306 benign) from patients in the United States was used in the training of our breast ultrasoundCADx system. A second sonographic database consisting of 456 lesions (145 malignant, 311 benign) was obtained from a separate patient population in Asia and used to test the trained classifier. Four sonographic features were extracted from each lesion (shape, margin sharpness, posterior acoustic behavior, and texture). These features were used to train a Bayesian neural network classifier. Both round‐robin and independent testing were used to evaluate the classifier with the two databases. Performance was assessed by calculating the area under the ROC curve (AUC) for each test. Results: The AUC of the independent test was 0.80 while the AUCs for the round‐robin tests were 0.87 and 0.88 for the Asian and American databases, respectively. The difference between the independent test AUC and the round‐robin AUC for the Asian database was statistically significant (p‐value=0.02). Conclusion: This work indicates that the breast ultrasoundCADx system is moderately robust across different patient populations. The statistically significant difference between the AUCs of the Asian database along with the similarity of the round‐robin AUCs indicate that while the sonographic features used by the system are useful in both databases, their relative importance differs. This motivates the future exploration of optimal feature sets to improve the overall performance of the CADx system. Conflict of Interest: Research supported in part by NIH. Some authors receive royalties, research funding, and/or are stock holders in Hologic.

An Effective Bayesian Neural Network Classifier with a Comparison Study to Support Vector Machine

We propose a new Bayesian neural network classifier, different from that commonly used in several respects, including the likelihood function, prior specification, and network structure. Under regularity conditions, we show that the decision boundary determined by the new classifier will converge to the true one. We also propose a systematic implementation for the new classifier. In our implementation, the tune of connection weights, the selection of hidden units, and the selection of input variables are unified by sampling from the joint posterior distribution of the network structure and connection weights. The numerical results show that the new classifier consistently outperforms the commonly used Bayesian neural network classifier and the support vector machine in terms of generalization performance. The reason for the inferiority of the commonly used Bayesian neural network classifier and the support vector machine is discussed at length.