Abstract
Examining Peripheral Blood Smear (PBS) images under a microscope remains the gold standard for diagnosing different types of anemia and evaluating Red Blood Cell (RBC) shape and size. However, manual checks are time-consuming and often depend heavily on the individual technician's experience. To address this, we developed a hybrid deep learning model enhanced with Explainable Artificial Intelligence (XAI) to analyze PBS images automatically. Our system uses a dual-path design—combining Convolutional Neural Networks (CNNs) for fine visual details like texture and edge shape with Vision Transformers (ViTs) for broader contextual information across the image. An adaptive attention mechanism then intelligently blends these two views. To make the system’s decisions clear to doctors, we integrated visual tools like Integrated Gradients and Grad-CAM. Tested on benchmark datasets, our model accurately identifies diverse types of anemia and abnormal RBC shapes, offering high diagnostic precision and transparent reasoning.
