Development of an Artificial Intelligence Method to Detect COVID-19 Pneumonia in Computed Tomography Images

Introduction: This study aimed to construct an artificial intelligence system to detect Coronavirus disease-2019 (COVID-19) pneumonia on computed tomography (CT) images and to test its diagnostic performance. Methods: Data were acquired between March 18-April 17, 2020. CT data of 269 reverse tran-scriptase-polymerase chain reaction proven patients were extracted, and 173 studies (122 for training, 51 testing) were finally used. Most typical lesions of COVID-19 pneumonia were la-beled by two radiologists using a custom tool to generate multiplanar ground-truth masks. Us-ing a patch size of 128x128 pixels, 18,255 axial, 71,458 coronal, and 72,721 sagittal patches were generated to train the datasets with the U-Net network. Lesions were extracted in the or-thogonal planes and filtered by lung segmentation. Sagittal and coronal predicted masks were reconverted to the axial plane and were merged into the intersect-ed axial mask using a voting scheme. Results: Based on the axial predicted masks, the sensitivity and specificity of the model were found as 91.4% and 99.9%, respectively. The total number of positive predictions has increased by 3.9% by the use of intersected predicted masks, whereas the total number of negative predic-tions has only slightly decreased by 0.01%. These changes have resulted in 91.5% sensitivity, 99.9% specificity, and 99.9% accuracy. Conclusion: This study has shown the reliability of the U-Net architecture in diagnosing typical pulmonary lesions of COVID-19 in CT images. It also showed a slightly favorable effect of the intersection method to increase the model's performance. Based on the performance level pre-sented, the model may be used in the rapid and accurate detection and characterization of the typical COVID-19 pneumonia to assist radiologists.