Evaluating the Diagnostic Accuracy of an Artificial Intelligence Tool for Pulmonary Abnormalities in Chest Radiographs: A Retrospective Study

Abstract

Introduction A retrospective, in-clinic study was conducted to validate the diagnostic performance of the Oxipit ChestEye™ AI tool in interpreting chest radiographs and to evaluate its impact on clinical outcomes and healthcare system efficiency. The performance of the AI model was compared with that of radiologists. Methods Chest radiographs from 1,470 patients, obtained from a public healthcare institution, were analyzed for nine pulmonary conditions: consolidation, heart enlargement, interstitial markers, linear atelectasis, no cardiopulmonary findings, nodule, pleural effusion, pneumothorax, and tuberculosis. Diagnostic accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 scores were calculated for Oxipit ChestEye™ and compared with radiologists' diagnoses, which served as the reference standard. Results The AI tool demonstrated accuracy ranging from 0.79 to 1.00 across the nine conditions evaluated. Sensitivity ranged from 0.25 to 1.00, and specificity ranged from 0.65 to 1.00. In differentiating between "Normal" and "Altered" categories, the AI model achieved an accuracy of 0.78. The F1 score for "Normal" was 0.86, whereas for "Altered," it was 0.52. Conclusions Oxipit ChestEye™ demonstrated high accuracy and efficiency in detecting and triaging pulmonary abnormalities in chest radiographs. While it is a useful tool to supplement the work of the radiologists, helping to reduce diagnostic time and improve patient care, it should not replace human radiologists.

Ähnliche Arbeiten