Deep Learning-Powered Classification of Thoracic Diseases in Chest X-Rays

TL;DR

This work tackles automated detection of thoracic diseases in chest X-rays, addressing diagnostic variability and severe class imbalance that challenge automated analysis. It employs transfer learning with ImageNet-pretrained AlexNet, ResNet152, and InceptionNet (V3), augmented by Focal Loss to emphasize hard, underrepresented cases, and uses Grad-CAM for interpretability. Evaluations on the NIH Chest X-ray14 dataset show meaningful gains after fine-tuning, with InceptionV3 delivering notable improvements in AUC and F1-Score and Grad-CAM heatmaps aligning model focus with clinically relevant regions. Although not yet state-of-the-art, the approach demonstrates potential to augment clinical workflows and provides a foundation for further optimization via domain-specific pre-training, advanced data augmentation, and efficient architectures.

Abstract

Chest X-rays play a pivotal role in diagnosing respiratory diseases such as pneumonia, tuberculosis, and COVID-19, which are prevalent and present unique diagnostic challenges due to overlapping visual features and variability in image quality. Severe class imbalance and the complexity of medical images hinder automated analysis. This study leverages deep learning techniques, including transfer learning on pre-trained models (AlexNet, ResNet, and InceptionNet), to enhance disease detection and classification. By fine-tuning these models and incorporating focal loss to address class imbalance, significant performance improvements were achieved. Grad-CAM visualizations further enhance model interpretability, providing insights into clinically relevant regions influencing predictions. The InceptionV3 model, for instance, achieved a 28% improvement in AUC and a 15% increase in F1-Score. These findings highlight the potential of deep learning to improve diagnostic workflows and support clinical decision-making.

Figures (6)

• Figure 1: Baseline performance metrics of the models.
• Figure 2: Evaluation results of models with pre-trained weights
• Figure 3: Training Loss for Epochs 1 to 20
• Figure 4: Figure 4. Activations of the different convolution layers: Early Convolution Layer (left), Middle Convolution Layer (middle), Final Convolution Layer (right). Grad-CAM visualizations highlight the regions of the chest X-ray most relevant to the model's predictions. The middle figure shows activations associated with Atelectasis, where the highlighted regions correspond to collapsed lung areas. This demonstrates the model's ability to focus on diagnostically significant features, aligning with clinical findings and validating its interpretability.
• Figure 5: Grad-CAM for AlexNet - Positive Atelectasis