Analysis of lung scan imaging using deep multi‐task learning structure for Covid‐19 disease. Issue 5 (3rd January 2023)

• Record Type: Journal Article
• Authors: Kordnoori, Shirin; Sabeti, Malihe; Mostafaei, Hamidreza; Banihashemi, Saeed Seyed Agha
• Is Part Of: IET image processing. Volume 17:Issue 5(2023)
• Page Start: 1534
• Page End: 1545
• Publication Date: 2023-01-03
• Subjects: image classification -- image segmentation; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/ipr2.12736
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract: Covid‐19 caused by the SARS‐CoV2 virus has become a pandemic all over the world. By growing in a number of cases, there is a need for clinical decision‐making system based on machine learning models. Most of the previous studies have examined only one task, while the detection and identification of infectious area are conducted simultaneously in the real world. Thus, the present study aims to propose a multi‐task model which can perform automatic classification‐segmentation for screening Covid‐19 pneumonia by using chest CT imaging. This model includes a common encoder for feature representation, one decoder for segmentation, and a multi‐layer perceptron for classification, respectively. The proposed model can evaluate three datasets, along with the effect of images size on the output of the model. The outputs were examined in both multi‐task and single‐task learning. The result indicates that the effect of multi‐task is significant in improving the results, which can increase the outputs of each task performance to 95.40% accuracy in classification and 95.40% in segmentation. Further, the model represented the highest results among the state‐of‐the‐art methods. The proposed model can be applied as a primary screening tool to help primary service staff in better referral of the suspected patients to specialists. Abstract : This paper presented a deep multi‐task model for screening COVID‐19 pneumonia using chest CT imaging.Tasks of the proposed model were segmentation and classification. Multi‐task learning improved these tasks outputs by increasing more than 7% in evaluation metrics.