Applications of Artificial Intelligence in Radiodiagnosis: Improving Accuracy in Medical Imaging

Abstract

Artificial intelligence in radiodiagnosis has emerged as a supportive approach for improving medical image interpretation, particularly in chest radiography, where multiple thoracic abnormalities may coexist. This study evaluated the application of artificial intelligence for multi-label classification of thoracic findings using the ChestMNIST repository dataset. A retrospective computational diagnostic accuracy design was used. The dataset included 112,120 grayscale chest radiographic images divided into training, validation, and test subsets, with 14 diagnostic labels including atelectasis, cardiomegaly, effusion, infiltration, mass, nodule, pneumonia, pneumothorax, consolidation, edema, emphysema, fibrosis, pleural thickening, and hernia. Images were normalized and analyzed using a lightweight one-vs-rest probabilistic artificial intelligence classifier, with performance assessed on the independent test set. Diagnostic evaluation included accuracy, area under the curve, sensitivity, specificity, precision, and F1-score. The model achieved a mean class-wise accuracy of 0.594 and a mean AUC of 0.584, with a mean sensitivity of 0.525 and specificity of 0.593. Class-wise performance varied, with stronger discrimination for edema, hernia, effusion, cardiomegaly, and consolidation. Low precision and exact-match accuracy reflected class imbalance and multi-label complexity. Artificial intelligence showed feasible radiodiagnostic application, but higher-resolution validation and radiologist-comparison studies remain necessary before clinical implementation.