Automated Detection and Segmentation of COVID-19 Infection using Machine Learning

S. K. Towfek^*1, Ehsaneh khodadadi², Fatma M. Talaat³

¹ Computer Science and Intelligent Systems Research Center, Blacksburg 24060, Virginia, USA

² Department of Chemistry and Biochemistry, University of Arkansas, Fayetteville, AR 72701, USA

³ Faculty of Artificial Intelligence, Kafrelsheikh University, Kafrelsheikh, Egypt

Emails: sktowfek@jcsis.org; ekhodada@uark.edu; fatma.nada@ai.kfs.edu.eg

Abstract

The accurate and timely segmentation of COVID-19 infection areas from CT scans is crucial for effective diagnosis and treatment planning. In this paper, we propose an automated approach utilizing machine learning techniques for COVID-19 infection segmentation. The proposed framework utilizes a convolutional neural network (CNN) architecture to extract informative features from CT scan images. These features are then fed into a segmentation model, which employs a combination of U-Net and attention mechanisms for accurate delineation of infection regions. To enhance the model's performance, we employ a transfer learning strategy by pretraining the CNN on a large dataset of general medical images. To evaluate the effectiveness of our approach, we conducted experiments on a diverse dataset consisting of CT scans from COVID-19 patients. The results demonstrate the superiority of our method in accurately segmenting infection areas, achieving an average Dice coefficient of 0.92 and a Jaccard index of 0.88. The proposed automated segmentation method offers significant potential for aiding radiologists and clinicians in identifying COVID-19 infection regions from CT scans rapidly and accurately. It can contribute to improved diagnosis, patient management, and treatment planning in the fight against the ongoing pandemic.

Keywords: Machine Learning (ML); COVID-19; Lung Segmentation; Computed Tomography;  Lesion Segmentation.