Volume 2
A Hybrid Diagnostic System to Detect COVID-19 Based on Selected Deep Features of Chest CT Images and SVM
Authors
Abdoljalil Addeh, Ali Hemmati, Ali Lari, Hafiz Mudassir Munir
Abstract
This paper proposes a hybrid diagnostic method for early detection of COVID-19 based on support vector machine (SVM) and selected deep features of chest computed tomography (CT) images. The developed method consists of four main parts including the feature extractor part, feature selection part, classifier part and optimizer part. In the feature extraction part, a convolutional neural network (ConvNet) is implemented for image preprocessing and extraction of new features from CT images. In the feature selection part, minimum Redundancy Maximum Relevance (mRMR) method is applied to select the most effective and informative features for extracted deep features by ConvNet. The selected features are fed into SVM in the classification part. Free hyper-parameters such as size and number of filters in ConvNet, and penalty factor in SVM control their accuracy and robustness. In the optimization part of the developed method, we applied the black widow optimization algorithm (BWOA) for optimal tuning of these parameters. The acquired outcomes demonstrated that the developed diagnostic method has excellent performance in the detection of COVID-19 and distinguishing it from other frequent respiratory illnesses using only small number of training data, which has huge possibility to help physicians and pulmonologist in performing a quick diagnosis. The developed diagnostic method can mitigate the enormous amount of work from professional treatment staff particularly when the healthcare system is overburdened.
Keyword: Early detection, ConvNet, COVID-19, Hybrid diagnostic method, Optimization algorithm, SVM.
PDF [ 1311.18 Kb ] | Endnote File | XML
