AI-Based Assessment of Chest Diseases on Chest X-ray Imaging Using DenseNet-121 and a CheXpert-Trained Model: A Comparative Diagnostic Accuracy Study

Chest X-rays are widely used to detect thoracic and lung conditions, but reviewing high volumes of radiographs can lead to workload strain and variation between interpreters. Artificial intelligence (AI), particularly deep learning neural networks like DenseNet-121, has shown strong potential to assist clinicians with automated image interpretation. However, AI models trained on large international datasets, such as CheXpert, may perform differently across distinct patient populations due to variations in imaging technique, patient demographics, and disease presentation.

The primary purpose of this study is to compare the diagnostic accuracy of a DenseNet-121 model trained or fine-tuned on local data against a DenseNet-121 model pretrained on the CheXpert dataset for identifying thoracic pathologies. Both models will evaluate de-identified frontal chest radiographs from adult patients. Model predictions will be compared against a reference standard established by expert radiologist consensus, with discordant findings resolved using chest computed tomography (CT). Findings will evaluate whether local model adaptation improves diagnostic precision and workflow efficiency in clinical settings.