A Full Framework For Automatic Detection of Acute Lymphoblastic Leukaemia

Galila, Mohamed A.; El-Hasnony, Ibrahim M.; Mostafa, Reham Reda; Barakat, Sherif

doi:10.21608/mjcis.2025.358215.1010

A Full Framework For Automatic Detection of Acute Lymphoblastic Leukaemia

Document Type : Original Research Articles.

Authors

Department of Information System, Faculty of Computer and Information System, Mansoura University.

10.21608/mjcis.2025.358215.1010

Abstract

Cancer is a common and serious disease manifesting itself in humans. Today, Acute Lymphoblastic Leukemia (ALL) is being dubbed as the most common type of cancer in several nations. It cannot be missed that most of the currently available methods for diagnosing ALL are quite traditional and not so sensitive and specific, and are also prone to missing minimal residual disease (MRD), thus risking a relapse. Furthermore, such methods take a lot of time and are not really efficient. In this context, we developed ALL-Predictor, a framework for early detection of ALL in microscopic images, which includes data augmentation techniques, preprocessing, segmentation using the Chan-Vese algorithm, morphological operations, feature extraction using the Gray Level Co-occurrence Matrix (GLCM), and data normalization
technique using Min-Max scaler. Recursive Feature Elimination (RFE) is used to select the features. Ultimately, Support Vector Machine (SVM) will conduct a final classification that indicates high confidence in performance. After Min-Max scaling, the accuracy of the ALL_DB1 dataset has increased from 88.42% to 98.42%, the accuracy of the ALL_DB2 dataset has increased from 84.84% to 98.29%, and the accuracy of the ALL_DB_OriginalSet set has increased from 79.50% to 96.46%. ALL-Predictor is well suited for providing accurate detection for the disease of ALL.

Keywords