Home-Based ABO Blood Group Classification from Microscopic Blood Sample Images Using Transfer Learning and Explainable AI

Accurate ABO blood group classification is essential for safe blood transfusions, organ transplants, and emergency medical care. Traditional serological methods are time-consuming, require specialized reagents, and are susceptible to human error, particularly in resource-limited settings. While deep learning has demonstrated remarkable success in various medical imaging tasks, its application to ABO blood group classification from microscopic images remains underexplored, and existing approaches lack the explainability necessary for clinical trust. In this paper, we propose a CNN-based framework for automated ABO blood group classification from microscopic blood sample images, leveraging transfer learning with MobileNetV2 and custom convolutional architectures. We employ CLAHE-based preprocessing, extensive data augmentation, and introduce Grad-CAM visual explanations to provide interpretable model decisions — enabling clinicians to understand which image regions drive classification outcomes. Experiments are conducted on the Barcelona PBC dataset and a custom ABO-labeled dataset, with models evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics. Our approach aims to bridge the gap between high-performance classification and clinical explainability, offering a faster, more transparent alternative to conventional blood typing methods. To the best of our knowledge, this is the first work to combine lightweight CNN architectures with Grad-CAM explainability specifically for ABO blood group classification.