An Explainable Deep – Learning based Multi-Label Image Classification for Chest X-Rays

Accurate diagnosis of respiratory diseases from chest X-rays, especially when multiple diseases coexist, is challenging. While deep learning has been effective in multi-label classification, its lack of interpretability hinders its adoption in critical medical applications. We purpose an explainable strategy utilizing three distinct techniques-Grad-CAM, LIME, and DeepLIFT—to offer comprehensive discernment into the model’s decision- making process. Our two-step training approach involved treating uncertain labels as zeros to establish baseline weights and then retraining with uncertain labels as ones led to significant performance gains, achieving an average AUC of 0.87 on the CheXpert dataset, surpassing existing methods (0.78-0.83). Our findings emphasize the need for comprehensive label analysis in medical imaging and demonstrate the effectiveness of specific explainability techniques in enhancing model interpretability for chest X-ray classification, ultimately aiding radiologists in clinical decision-making.