Comparison of Interpretations of Chest Radiographs: Case of Interpretation by Radiologist and Artificial Intelligence Technology, Kalasin Hospital

Abstract

          The amount of chest radiograph (CXR) is continuously increasing, resulting increase the workload of radiologist. Artificial intelligence (AI) technology has been to assist in the interpretation of CXR for reduce steps and increase work efficiency. AI technology is continuously being develop to achieve higher accuracy. Objective: To study the comparison of CXR interpretation by AI technology and radiologist. Methods: Retrospective descriptive study, a samples of 2,207 CXR images in Kalasin hospital with interpreting by AI technology and by radiologist. Comparative analysis of the interpreting of CXR by AI technology and radiologists. Results: The interpreting by AI technology and radiologists is mostly highly consistent with a Kappa of 0.67 to 1.0, and the sensitivity and specificity are similar. Some pathologies such as nodules and calcified nodules have lower sensitivity and consistency. As for fracture bone, it is not indicated by AI technology, but indicate by radiologist. There were inconsistencies in some lesions such as infiltration, nodule and some locations such as both lower lungs, hilar and retrocardiac regions. Conclusions: The interpretation of CXR by AI technology and radiologists had a high to very high level of consistency. Although fracture bone was not indicated by AI ​​interpretation which was different from the interpretation by radiologists. The interpretation by AI technology will have lower sensitivity such as nodule or calcified nodule and some locations such as hilar and retrocardiac regions that require caution ininterpretation. However AI technology can be applied to the interpretation of CXR images to reduces work steps and waiting times