A Hybrid Segmentation and Deep Learning Framework for Automated Leukemia Detection in Blood Smear Images

Leukemia is a life-threatening hematological malignancy in which rapid and accu rate diagnosis is essential for effective clinical intervention. Conventional manual examination of peripheral blood smear images is labor-intensive, time-consuming, and highly dependent on the expertise of hematologists, leading to potential vari ability in diagnosis. To address these limitations, this study proposes a hybrid automated framework for leukemia detection that integrates advanced image segmentation and deep learning-based classification. The proposed approach consists of a dual-stage pipeline. In the first stage, a hybrid segmentation method combining K-means clustering and region growing is employed to accurately isolate white blood cells and extract cellular regions of interest from blood smear images. This hybrid technique leverages the strengths of both unsupervised clustering and spatial connectivity to improve segmentation precision. In the second stage, the segmented cell regions are classified using a pre-trained MobileNetV2 convolutional neural network. Experimental results demonstrate that the proposed framework achieves high segmentation quality and classification performance, confirming its effective ness and robustness. The model shows strong accuracy in detecting leukemia, highlighting its potential as a reliable computer-aided diagnostic tool.