Artificial Intelligence with based Chest X-ray Image Analysis Screening 2019 Novel Coronavirus Pneumonia (COVID-19)

micr-087-prapasri-benchasiriluck
Prapasri Benchasiriluck, Wasun Chantratita, Sathit Niramitmahapanya, Tarit Taerakul, Krisna Dissaneevate, Thouantosaporn Suwanjutah, Ronasit Poomma, Arakin Rakchittapoke, Sonchai Seangtian, and Peerapong Modethes

Abstract

The objectives of this experimental research were to study the usage of artificial intelligence (AI) in chest X-ray (CXR) analysis of pneumonia, which has been proven to be significantly accurate and effective in rapid screening for coronavirus disease 2019 (SARS-CoV-2). The study revealed no difference in results when comparing CXR analysis between AI and radiologists. Furthermore, AI intelligence is also compatible with complex diagnosis, such as the fifteen types of lung diseases. Samples were calculated using W.G. Cochran’s formula, while data collection was done using probabilistic sampling techniques, which produce a 95% confidence interval for an unknown population. At first, 10,000 samples were collected, resulting in a 1% error. To minimize the potential error, 12,881 samples were collected instead and divided into three phases: 210 cases, 1,100 cases, and 11,571 cases, respectively. According to results, the prognosis of pneumonia’s classification models was revealed by a dataset of AUC and ROC scores (area under the receptor functional curve: AUC). Model 1 and Model 2 show different types of histogram distribution values, which are 0.97 and 0.96, respectively. And the combined model got 0.98 AUC scores, which seem to have a better performance than each model in action by having a value of 0.98 AUC. During Phase 1, some significant values were revealed: high-performance AUC of 0.9878, sensitivity of 0.9717, and specificity of 0.9198. In Phase 2, data were obtained from comparative tests between AI and radiologists, which are 94% accurate. In Phase 3, data were collected from 11,571 samples. CXR images were analyzed by AI. 1,628 images were shown as positive for pneumonia detection (14%), and 9,943 images were shown as negative (86%). In addition, by using real-time RT-PCR examination, 19 samples were detected with Coronavirus disease 2019 infection (0.16%), and 11,552 samples were undetected (99.84%). Throughout the 3 phases of this research, pneumonia and pulmonary tuberculosis were found, and their severity was accelerating. There were 10 positive samples from nasopharyngeal swab tests for Coronavirus disease 2019 infection, while others (9 in number) were inconsistent; this could possibly be related to new variants of Coronavirus disease 2019 that are currently spreading. Therefore, it’s necessary to make sure this study could go further to provide an effective and accurate result that would allow patients to receive proper treatment in time.

Published on: April 20, 2023
doi: 10.17756/micr.2023-087
Citation:  Benchasiriluck P, Chantratita W, Niramitmahapanya S, Taerakul T, Dissaneevate K, et al. 2023. Artificial Intelligence with based Chest X-ray Image Analysis Screening 2019 Novel Coronavirus Pneumonia (COVID-19). J Med Imaging Case Rep 7(1): 5-15.

304
Downloads