Automated detection of COVID-19 infected lesion on computed tomography images using faster-RCNNs

firdaus, firdaus (2020) Automated detection of COVID-19 infected lesion on computed tomography images using faster-RCNNs. Engineering Letters, 28 (EL_28_). pp. 1295-1301.

[thumbnail of 19.pdf] Text
19.pdf - Published Version

Download (1MB)

Abstract

The gold standard of a definitive test for the 2019 novel Corona Virus (SARS-CoV-2) is reverse-transcription polymerase chain reaction (RT-PCR). However, its sensitivity ranged between 50%-90% with high false negatives. Currently, false negatives are real clinical problems, caused by the absence of antibodies formation during sampling (incubation period), impaired antibody formation in immunocompromised patients, apart from sample acquirement technique and transportation issue. Thus, repeated RT-PCR testing is often needed at the early stage of the disease, which may prove to be difficult in a pandemic situation. In some research, the chest computed tomography (CT) image was a rapid and reliable method to diagnose patients with suspected SARS-CoV-2 with higher sensitivity compared to RT-PCR test, particularly the lab test is negative. In this study, 420 CT images with 2,697 features from seven patients infected by SARS-CoV-2 and 200 CT images from healthy individuals are used for analyzing. The convolutional neural networks (CNNs) with Faster-RCNNs architecture is proposed to process the infected lesion detection. As a result, the proposed model shows 90.41% mAP, 99% accuracy, 98% sensitivity, 100% specificity, and 100% precision of classifier performances. All performance produces a 100% score when it tests on external data CT image. It can be seen from the detection result that Ground-glass opacities (GGO)-principal lesions on CT images in the peripheral and posterior sections of the lungs should be strongly suspected of developing SARS-CoV-2 pneumonia. On average, it took less than 0.3 seconds per image to detect the abnormalities from a CT image from data pre-processing to the output of the report. For a frontline clinical doctor, the proposed model may be a promising, supplementary diagnostic process. © 2020, International Association of Engineers. All rights reserved.

Item Type: Article
Subjects: Q Science > Q Science (General) > Q334-342 Computer science. Artificial intelligence. Algorithms. Robotics. Automation.
Divisions: 09-Faculty of Computer Science > 56201-Computer Systems (S1)
Depositing User: Mr Firdaus Firdaus
Date Deposited: 17 Mar 2023 15:13
Last Modified: 17 Mar 2023 15:13
URI: http://repository.unsri.ac.id/id/eprint/90694

Actions (login required)

View Item View Item