Explainable Deep Learning for Automated Classification of Acute Lymphoblastic Leukemia
A. L. Akshay, Hari Krishnan N, Tricha Anjali · 2025
Acute lymphoblastic leukemia (ALL) is a cancer that spreads quickly and can be fatal; it needs to be diagnosed as soon as possible. This study presents a deep learning approach for automatic ALL classification using the Swin Transformer model with an impressive 96 % accuracy rate. Using hierarchical selfattention mechanisms, the Swin Transformer effectively captures both local and global image features, allowing for accurate classification. Score-CAM is used to improve interpretability by visualizing model attention and providing clinicians with information about decision-making. Weighted sampling, focal loss for handling imbalances, and data augmentation all improve the performance of the model. Evaluation metrics show consistent, high accuracy across classes, such as the classification report and confusion matrix. Our work provides a dependable, comprehensible, and scalable approach to automated leukemia classification, bridging the gap between AI-driven leukemia diagnosis and clinical application.