Classification and Region Analysis of COVID-19 Infection Using Lung CT Images and Deep Convolutional Neural Networks

40 Pages Posted: 8 Feb 2022

See all articles by Saddam Hussain Khan

Saddam Hussain Khan

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab

Anabia Sohail

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab

Asifullah Khan

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab

Yeon Soo Lee

Catholic University of Daegu

Multiple version iconThere are 2 versions of this paper

Abstract

COVID-19 is a global health problem that requires efficient diagnostic techniques to detect and thus limit its spread. However, the accurate analysis of COVID-19 infected regions is challenging due to (i) minor contrast variation between infected and the background boundaries, (ii) high texture variation within homogeneous infected regions. In this regard, a new two-stage deep Convolutional Neural Networks (CNNs) based framework is proposed for demarcating COVID-19 infected regions in Lung CT images. In the first stage, COVID-19 specific image features are extracted using a two-level discrete wavelet transformation and then classified using the proposed deep CoV-CTNet. A new CNN block named “CoV-CTNet” is developed that systematically implements region and edge operations to capture features related to the COVID-19 lungs-infection specific patterns such as homogeneous region, texture variation, and boundary contrast. On the other hand, in the second stage, the infectious CT images are provided to the segmentation models to identify and analyze COVID-19 infectious regions. In this regard, a novel semantic segmentation model CoV-RASeg is proposed, which employs average and max-pooling operations in the encoder and decoder blocks. This systematic utilization of average and max-pooling operations helps the proposed CoV-RASeg learn the region homogeneity and boundaries related patterns simultaneously. Moreover, the idea of attention block is incorporated in CoV-RASeg and customized segmentation models to deal with mildly infected regions. The performance of the proposed CoV-CTNet is evaluated using accuracy (98.80%), MCC (0.98). While the proposed SA-CoV-RASeg achieves an IoU score of 98.75% and Dice Similarity score of 0.954. The promising results suggest that the proposed framework has the potential to help the radiologists in the identification and analysis of COVID-19 infected regions. Source code and related GUI details are provided at https://github.com/PRLAB21/COVID-19-Diagnostic-System.

Note:
Funding: This work was conducted with the support of the research grant of National Research Foundation of Korea (2017R1A2B2005065). This study was also supported by the PIEAS IT endowment fund and HEC indigenous Scholarship under the Pakistan Higher Education Commission (HEC).

Declaration of Interests: Authors declared no conflict of interest.

Keywords: COVID-19, CT image, Convolutional Neural Networks, Transfer learning, Classification, and Segmentation

Suggested Citation

Khan, Saddam Hussain and Sohail, Anabia and Khan, Asifullah and Lee, Yeon Soo, Classification and Region Analysis of COVID-19 Infection Using Lung CT Images and Deep Convolutional Neural Networks. Available at SSRN: https://ssrn.com/abstract=4029649 or http://dx.doi.org/10.2139/ssrn.4029649

Saddam Hussain Khan

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab ( email )

Islamabad
Pakistan

Anabia Sohail

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab ( email )

Islamabad
Pakistan

Asifullah Khan (Contact Author)

Pakistan Institute of Engineering and Applied Sciences (PIEAS) - Pattern Recognition Lab ( email )

Islamabad
Pakistan

Yeon Soo Lee

Catholic University of Daegu ( email )

Korea, Republic of (South Korea)

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
17
Abstract Views
217
PlumX Metrics