A dynamic pooling based convolutional neural network approach to detect chronic kidney disease. (September 2020)

• Record Type: Journal Article
• Authors: Navaneeth, Bhaskar; Suchetha, M
• Is Part Of: Biomedical signal processing and control. Volume 62(2020)
• Publication Date: 2020-09
• Subjects: Chronic kidney disease -- Convolutional neural network -- Dynamic pooling -- Feature selection -- Support vector machine; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102068
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: This work explores the use of salivary urea as a potential biomarker to detect CKD noninvasively. A novel sensing module is designed and implemented for detecting salivary urea concentration. We have implemented a 1-D CNN-SVM hybrid network with a dynamic pooling approach and a feature pruning algorithm for automated classification and prediction. The performance evaluation of the proposed hybrid network is carried out, and it is compared with the conventional data classification methods. Abstract: Objective: In this paper, we present a deep learning technique and a novel detection methodology to detect Chronic Kidney Disease (CKD) from saliva samples. Methods: A hybrid deep learning network comprising of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) classifier is introduced to overcome the challenges faced by conventional data classification networks. A novel dynamic pooling approach and a feature pruning algorithm are introduced in the network to select the most relevant features for the classification operation. We have examined the concentration of urea in the saliva sample to detect the disease. A new detection module is developed for testing the samples. Results: The CNN-SVM network achieved an average accuracy of 97.67% and a sensitivity and specificity of 97.5% and 97.83%, respectively. The conventional CNN model achieved an average accuracy of 96.51%. We have compared our proposed model with other existing algorithms, and it is observed that the performance achieved by this model is higher than other well-known data classification methods. Conclusion: Combining CNN with the SVM classifier enables the network to analyze the sensor data to make predictions more accurately. The use of dynamic pooling and feature pruning algorithm significantly improved the prediction accuracy of the network. The experimental results show that the proposed method provides acceptable classification accuracy and has the potential to be implemented in clinical practice. Significance: Our study result shows that the proposed methodology can be used for detecting CKD non-invasively. The proposed deep learning network provides accurate predictions compared to other data classification methods.