An Effective Model for Thoracic Disease Diagnosis in Diverse Healthcare Environments

Thoracic diseases, affecting the lungs and the heart, are the major causes of global illness and mortality. A reliable diagnosis of these conditions is crucial for timely intervention and treatment. However, existing deep learning models struggle with substantial issues concerning computational complexity and scalability due to their reliance on anatomical-prior maps. Our study aims to develop an efficient deep-learning model to enhance classification accuracy and generalization of thoracic disease diagnosis while maintaining computational efficiency. We present Progressive Residual Network 50 (ProResNet50), a Convolutional Neural Network (CNN) architecture for thoracic disease classification. In contrast to previous research works that use various self-attention techniques and chest ROI masks incorporating probabilistic anatomical prior information, the proposed method uses patterning cluster-distributed features and feature-wise weighted aggregation, which maps and emphasizes important areas of an input image, improving the model’s potentiality to focus on critical features, and minimizes computational complexity while retaining high feature representation quality. The proposed model excels over existing techniques for medical illness classification on the dataset named NIH ChestX-ray14, achieving an AUC of 94.66%. The ProResNet50 model demonstrates its effectiveness for thoracic disease classification tasks with 90% accuracy. The model's efficiency makes it a strong candidate for widespread clinical deployment, enhancing thoracic disease diagnosis.