Deep Learning Approaches for COVID - 19 Diagnosis

Chetan Sagarnal, Shridhar B. Devamane, Ravi Hosamani, Trupthi Rao
2021 International Workshop on Informatics & Data-Driven Medicine  
Coronavirus disease (Covid19) is a pandemic communicable disease that has a serious risk of speedy transmission. Identifying and isolating the affected person is the initiative mark to counter this virus. In regard to this matter, chest radiology images have been manifested to be a powerful screening approach of Covid19 positive patients. Many Artificial Intelligence based solutions have evolved for fast screening of radiological images and more precise in detecting Coronavirus disease. To make
more » ... the proposed model more powerful, labeled chest Xray datasets comprising two categories Covid19 and Non-Covid from kaggle uci repository data set are used in this work. To perform feature extraction, effective CNN structures, namely EfficientNet, VGG-16 and Densenet-121 with ImageNet pre-training weights are applied. The features produced are moved to custom fine-tuned top layers which are then followed by a group of model snapshots. In this study, the main objectives are to create database of Covid19 patients and to develop different Deep learning model for analysis of Covid19 pneumonia and then to train the deep learning models to get desired accuracy. A deep learning-based approach using Densenet-121 with ReLu activation function is proposed to effectively detect Covid19 patients X-ray images. The model is trained on Covid19 dataset which consisted of 2159 labelled X-ray images (576 images are of confirmed Covid19 patients and 1583 are of non-covid patients) and achieved overall accuracy of 95.04% in classifying the X-ray images and tested this model on Covid dataset containing 25 unidentified chest X-ray images. As a final step, we performed two-class classification of unidentified X-ray images as Covid and Normal using the proposed deep learning model.
dblp:conf/iddm/SagarnalDHR21 fatcat:63yipxsgxrg2logen2j4uyx2ji