YOLO-BASED MULTI-CLASS AND SEVERITY LEVEL CLASSIFICATION OF BLOOD CANCER FROM MICROSCOPIC SMEAR IMAGES

– Acute Lymphoblastic Leukemia (ALL) and other hematological malignancies remain significant causes of mortality worldwide, necessitating rapid and precise diagnostic interventions. Manual examination of peripheral blood smear (PBS) images is labor-intensive, time-consuming, and prone to human error. This research proposes an automated, high-precision framework using the YOLO (You Only Look Once) architecture-specifically YOLOv8/v11- to simultaneously perform multi-class cell identification and severity level grading from microscopic smear images. he proposed system utilizes a dual-task deep learning approach. First, the model is trained on a diversified dataset of microscopic images (e.g., ALL-IDB, C-NMC) to identify multiple cell classes, including lymphoblasts, neutrophils, and eosinophils. Second, the framework incorporates a hierarchical classification layer to grade the severity of the malignancy (e.g., Early Pre-B, Pre-B, and Pro-B stages) based on morphological features such as nuclear-cytoplasmic ratio, chromatin density, and cellular irregularities. To enhance performance in dense cell environments, integrated Attention Mechanisms (CBAM) and Multi-Scale Feature Fusion into the YOLO backbone, ensuring the detection of small and overlapping blastic cells.