Optimal Ensemble learning model for COVID-19 detection using chest X-ray images. (March 2023)

• Record Type: Journal Article
• Authors: Balasubramaniam, S; Satheesh Kumar, K
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: COVID-19 -- Preprocessing -- Feature Extraction -- Classification -- Optimization; 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.2022.104392
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Proposes a Self Adaptive Kill Herd Optimization (SAKHO) Algorithm for training the optimized NN model via selecting the optimal weights. Introduces an ensemble classifier that includes SVM, RF, CNN, and optimized NN for final classification using the COVID-19 chest images. The performance of proposed ensemble + SAKHO model based COVID-19 detection by X-ray images is better than the standard conventional models like FM-HCF-DLF, DBN, FCM, LSTM, ensemble + KHO, ensemble + EHO, ensemble + WOA, and ensemble + BOA, respectively. Abstract: COVID-19 pandemic is the main outbreak in the world, which has shown a bad impact on people's lives in more than 150 countries. The major steps in fighting COVID-19 are identifying the affected patients as early as possible and locating them with special care. Images from radiology and radiography are among the most effective tools for determining a patient's ailment. Recent studies have shown detailed abnormalities of affected patients with COVID-19 in the chest radiograms. The purpose of this work is to present a COVID-19 detection system with three key steps: "(i) preprocessing, (ii) Feature extraction, (iii) Classification." Originally, the input image is given to the preprocessing step as its input, extracting the deep features and texture features from the preprocessed image. Particularly, it extracts the deep features by inceptionv3. Then, the features like proposed Local Vector Patterns (LVP) and Local Binary Pattern (LBP) are extracted from the preprocessed image. Moreover, the extracted features are subjected to the proposed ensemble model based classification phase, including Support Vector Machine (SVM), Convolutional Neural Network (CNN), Optimized Neural Network (NN), and Random Forest (RF). A novel Self Adaptive Kill Herd Optimization (SAKHO) approach is used to properly tune the weight of NN to improve classification accuracy and precision. The performance of the proposed method is then compared to the performance of the conventional approaches using a variety of metrics, including recall, FNR, MCC, FDR, Thread score, FPR, precision, FOR, accuracy, specificity, NPV, FMS, and sensitivity, accordingly.