Analysis of Chest Radiographs Using Deep Learning: A Multi-Model Approach for Detection of Thoracic Abnormalities

Abstract

The increasing utilization of chest radiographs in medical diagnostics necessitates automated systems that can assist radiologists in accurate and efficient interpretation. This paper presents a comprehensive deep learning-based framework for automated analysis of chest X-rays, focusing on three critical diagnostic tasks: abnormal rib count detection, cardiomegaly identification, and pneumonia classification. Our multi-model ap- proach employs specialized convolutional neural network (CNN) modules including a custom lightweight regression model for rib counting achieving Mean Absolute Error (MAE) of 0.58 ribs with 89% accuracy within ±1 rib prediction, transfer learning-based architectures for pneumonia detection with 94.87% accuracy us- ing ResNet50, and cardiomegaly detection through cardiothoracic ratio analysis. The proposed framework addresses significant gaps in automated chest radiograph analysis by providing simul- taneous assessment of structural abnormalities, cardiac condi- tions, and pathological conditions. Evaluated on multiple datasets including VinDr-RibCXR, pediatric pneumonia dataset, and NIH-derived cardiomegaly subset, our system demonstrates significant potential for deployment in resource-constrained healthcare environments, offering rapid diagnostic assistance while maintaining high sensitivity and specificity rates across all three diagnostic modules.