AI Powered Leukemia Screening – Hybrid Deep Learning with SHAP and LIME Insights

Leukemia, a hematological malignancy, requires early and accurate diagnosis to improve patient outcomes. This study proposes a novel hybrid deep learning model combining a Convolutional Neural Network (CNN) encoder with a Vision Transformer (ViT) for classifying leukemia from microscopic blood cell images. Utilizing the C-NMC Leukemia dataset, we preprocess images with Contrast Limited Adaptive Histogram Equalization (CLAHE) and apply data augmentation to enhance robustness. A grid search over learning rates, optimizers, and batch sizes identifies optimal hyperparameters, achieving a maximum validation accuracy of 85.61%. Explainable AI (XAI) techniques, including SHAP and LIME, provide insights into the model’s decision-making process, highlighting critical image regions for classification. The proposed model demonstrates competitive performance, with potential for clinical deployment in automated leukemia screening.