A Novel Concavity-Based Segmentation and ANN for the Automatic Detection and Classification of Leukemia

Acute Lymphoblastic Leukemia (ALL) is a severe hematological disorder that primarily affects white blood cells (WBCs) and necessitates immediate clinical intervention. Conventional diagnostic methods rely heavily on manual microscopic examination, which is labor-intensive and subject to observer variability. This study proposes an automated system for ALL detection by integrating Image Processing (IP) techniques with Artificial Neural Networks (ANNs). The nuclei of WBCs were segmented using three approaches: thresholding, color-based K-means clustering, and a concavity-based method. Among these, the concavity-based segmentation achieved superior results, with Dice coefficient values ranging between 0.8 and 1.0. From the segmented nuclei, 18 statistical features encompassing geometrical, textural, and fractal characteristics were extracted. Significant features were identified using the T-test and employed as inputs to a three-layer Backpropagation Neural Network (BPNN) for the classification of blood smear images (BSIs) into malignant and benign categories. The proposed framework was evaluated on the publicly available ALL-IDB2 dataset. Experimental results demonstrated that the ANN achieved its highest performance when fractal features were used as input, with five hidden neurons. Under these conditions, the model yielded a sensitivity of 95%, a specificity of 100%, a positive predictive value (PPV) of 100%, a negative predictive value (NPV) of 95%, and an overall accuracy of 97.43%. These results indicate that the proposed ANN-based system, combined with robust feature extraction and selection, provides an efficient and reliable tool for the automated detection of ALL. The findings highlight the potential of integrating statistical image features with machine learning models to improve the speed and accuracy of leukemia diagnosis.