Metaheuristic Deep Learning Models for Leukemia Classification and Grading

Abstract: Hematological malignancies, specifically Leukemia, manifest through abnormal white cell proliferation in the bone marrow. Diagnosing this quickly is key for survival. However, looking at slides manually is slow and errors occur. This study works on a dual-stage framework. It couples Particle Swarm Optimization (PSO) with a ResNet-18 back- bone. The architecture handles multi-class classification (ALL, AML, CLL, CML) and severity grading (Grades 1-3) at the same time. PSO functionality is used for hyperparameter tuning. This happens before feature extraction. Validation metrics indicate a precision maximum of 94.2%.

Keywords: Leukemia, Deep Learning, ResNet-18, PSO, Classification, Grading.