Evaluation of Heatmaps as an Explicative Method for Classifying Acute Lymphoblastic Leukemia Cells

Explainable artificial intelligence (XAI) is a field of research that has captured researchers' interest in recent years. These algorithms seek to bring transparency to artificial intelligence (AI) models, which are often highly complex and opaque algorithms. Implementing explanatory methods like heat maps to pathological diagnosis systems opens the door to using AI in potentially life-saving applications, generating new tools as auxiliaries in the corroboration of predictions made by an AI. In the present work, retraining four CNN models (VGG16, VGG19, ResNet50, and MobileNet V1) was performed, performing a fine tuning to classify segmented images without a background of Acute Lymphoblastic Leukemia (ALL) cells. Heat maps were generated using the iNNvestigate library, selecting the LRP, Deep Taylor, and Input*Gradient methods for this end. With the help of five hematologists and experts in morphological cell classification, the 120 generated heat maps were evaluated. The evaluation focused on the amount of information provided by the heat maps and how they relate to morphological characteristics present in the classified cells. Results of the best heatmaps and hematologist evaluations are presented in this work. The central outcome is that the heatmaps must include morphological information to be a valuable tool for medical diagnosis systems.