Lightweight and Efficient YOLOv8 With Residual Attention Mechanism for Precise Leukemia Detection and Classification

Abstract

Leukemia, defined by the abnormal growth of white blood cells, poses diagnostic difficulties due to its diverse symptoms and swift progression. Timely and precise detection is vital for effective treatment and better patient outcomes. This paper introduces a novel lightweight YOLOv8 model, integrated with a residual attention mechanism, aimed at improving leukemia detection and classification. Enhancements to the YOLOv8n architecture include Depthwise Separable Convolution (DWSCNN) and Residual Convolution Block Attention Mechanism (RCBAM) layers, which strengthen feature extraction and contextual information gathering. Trained on a comprehensive dataset of blood cell images annotated for various leukemia stages: benign, malignant-early, malignant-pre, and malignant-pro, the model employs noteworthy results, achieving the mAP of 98.4%, F1-score of 96.2%, and an inference speed of 3.5 milliseconds, significantly surpassing traditional YOLOv8 variants and other leading techniques. The proposed model not only improves diagnostic precision but also minimizes computational requirements, making it suitable for use in clinical settings, especially where resources are limited. By enabling early and precise detection of leukemia, this model holds promise for advancing treatment strategies and improving patient outcomes, paving the way for future innovations in medical imaging and automated disease diagnosis.