Transfer Learning Paradigm: A Comparative Analysis of VGG16 and EfficientNet in Childhood Leukemia Detection

Abstract

Childhood leukemia, a formidable health challenge, demands innovative strategies for timely detection and intervention. This study leverages the formidable capabilities of cutting-edge deep learning models, VGG16 and EfficientNetB3, to intricately classify a comprehensive dataset comprising 15,135 cell images from 118 patients. Resultantly, VGG16 achieves a commendable classification accuracy of 77%, while the EfficientNetB3 model excels with an exceptional 91% accuracy. Beyond classification proficiency, this research underscores the urgency of early detection in childhood leukemia, shedding light on the transformative potential of deep learning models in enhancing diagnostic capabilities. The findings not only pave the way for refined classification methodologies but also illuminate promising avenues for timely, personalized, and targeted therapeutic interventions. This holistic approach holds promise for significantly improving outcomes and quality of life for young leukemia patients, emphasizing the indispensable role of technology in propelling advancements in pediatric oncology.