Leukemia Diagnosis using Transfer Learning: An Efficient Approach

Abstract

Leukemia, the most prevalent type of blood cancer, affects both adults and children. For effective intervention, this necessitates prompt detection. Nevertheless, the conventional manual diagnostic techniques have lengthy procedures and are subject to variances based on skill. We use transfer learning techniques in our work, which take advantage of knowledge from models trained on massively parallel datasets. This method greatly improves time and cost efficiency by lowering the quantity of labeled data and processing resources required for the particular task of identifying acute lymphoblastic leukemia (ALL). This paper presents an automated leukemia diagnostic approach that makes use of machine learning methods, such as transfer learning and deep learning-based convolutional neural networks (CNNs). We present a approach that integrates CNN layers within Transfer Learning Architectures. The model that used EfficientNetB3 performed the best out of all the CNN architectures that were previously tested on the C-NMC-2019 ALL Dataset. It had 100% training accuracy, 96.87% testing accuracy, 96.9% F1-Score, 96.24% recall, and 97.58% precision. This model is considered one of the most promising models. The solution under proposal tackles the critical requirement of diagnosing leukemia early. By effectively examining microscopic images, identifying crucial information, and using filtering algorithms to improve accuracy, it overcomes the shortcomings of manual methods. By giving physicians and other healthcare professionals an accurate tool, this automated technique promises to increase blood cancer identification and greatly improve patient care and management of ALL.