Explainable AI for Leukemia Diagnosis: Modified ResNet with Hard Attention Mechanisms

Leukemia is a highly complex and aggressive form of cancer, necessitating accurate and timely diagnosis for effective treatment. This study investigates the use of Explainable Artificial Intelligence (XAI) techniques to enhance computer-aided diagnosis (CAD) systems for leukemia. A modified ResNet architecture, integrated with a hard attention mechanism, is proposed to improve model performance and interpretability. The approach emphasizes robust segmentation of white blood cell (WBC) nuclei, utilizing advanced image processing techniques and U-Net models for precise feature extraction. Model evaluation focuses on both accuracy and interpretability, employing Grad-CAM and clustering space analysis to assess network focus and transparency in decision-making. The results underscore the strengths and limitations of the proposed architecture, offering insights into further advancements in XAI to improve the reliability and transparency of CAD systems. This work ultimately aims to advance clinical outcomes in leukemia diagnosis by providing more reliable and interpretable diagnostic tools. The proposed model produces 99.7% of accuracy for detection and classification of leukemia.