Explainable Deep Learning for Lung Disease Detection on Chest X-ray Images Using Local Interpretable Model-Agnostic Explanations (LIME)

Artificial intelligence models, especially deep learning, are increasingly used for medical image analysis but often operate as black boxes, limiting trust and adoption in critical domains such as healthcare. This study develops a ResNet18-based convolutional neural network to classify chest X-ray images into three categories: normal, COVID-19, and pneumonia, using a dataset of 3,009 images. The model is trained from scratch with tailored preprocessing, architectural modifications, regularization, and learning rate scheduling to enhance generalization. Performance evaluation shows high effectiveness, with 97% precision, recall, and F1-score, and 98% overall accuracy. To address the interpretability gap, the Local Interpretable Model-agnostic Explanations (LIME) method is applied to highlight image regions most influential for each prediction. The resulting visual explanations help distinguish among the three lung conditions and demonstrate that integrating XAI methods such as LIME with deep learning can provide clinically relevant interpretability for chest X-ray classification.