Thoramo: An AI-Powered System for Automated Detection and Prioritization of Thoracic Diseases from Chest X-rays

Chest X-rays are essential for diagnosing thoracic diseases, but their interpretation remains time-consuming and challenging for pul-monologists, particularly in high-volume clinical settings. This paper presents Thoramo, an AI-powered diagnostic system utilizing a 121-layer Dense Convolutional Network (DenseNet) trained on 112,120 labeled chest X-rays to detect 14 thoracic conditions, achieving AUROC scores ranging from 0.7345 to 0.9371. Thoramo features a user-friendly interface that enables manual X-ray uploads and immediate analysis, prioritizing scans based on critical findings rather than a traditional first-come, first-served approach. The system incorporates Class Activation Mapping (CAM) to visualize areas of concern, enhancing interpretability for clinical decision-making. An evaluation conducted by Dr. Khaled Alobaidellah from Sulaiman Al Habib Hospital, Riyadh, emphasized its effectiveness in managing critical diseases and reducing delays in diagnosis. The study also highlights ethical and technical limitations, including concerns about data privacy, algorithmic biases, and dataset robustness. Thoramo represents a step towards integrating AI into pulmonology workflows, with future research focusing on addressing these challenges to ensure reliable clinical adoption. Figure 1. Thoramo, a 121-layer CNN, processes a chest X-ray, predicts pneumonia, and highlights key areas indicative of the pathology.