Automated leukemia detection in blood microscopic images using statistical texture analysis

TLDR

Pathological image analysis and quantitative microscopy are essential for accurate diagnosis of serious diseases such as leukemia, hepatitis, AIDS, and psoriasis, with acute lymphocytic leukemia characterized by lymphoblast accumulation in the bone marrow. The study proposes a texture‑based method for automated detection of leukemia. The method extracts nucleus texture and shape features, segments white blood cells via K‑means clustering, and uses these features to classify cells as normal lymphocytes or lymphoblasts for leukemia detection. The approach was evaluated on 108 blood smear images and its performance was validated against a hematologist’s assessment.

Abstract

Pathological image analysis plays a significant role in effective disease diagnostics. Quantitative microscopy has supplemented clinicians with accurate results for diagnosis of dreaded diseases such as leukemia, hepatitis, AIDS, psoriasis. In this paper we present a texture based approach for automated leukemia detection. Acute lymphocytic leukemia (ALL) is a malignant disease characterized by the accumulation of lymphoblast in the bone marrow. Texture features of the blood nucleus are investigated for diagnostic prediction of ALL. Other shape features are also extracted to classify a lymphocytic cell in the blood image into normal lymphocyte or lymphoblast (blasts). Initial segmentation is done using K-means clustering which segregates leukocytes or white blood cells (WBC) from other blood components i.e. erythrocytes and platelets. The results of K-means are used for evaluating individual cell shape, texture and other features for final detection of leukemia. A total of 108 blood smear images were considered for feature extraction and final performance evaluation is validated with the results of a hematologist.