FUSION BASED IDENTIFICATION AND CLASSIFICATION MODEL FOR β - THALASSEMIA DISEASE STAGES

Abstract

β-Thalassemia is an autosomal recessive haemoglobinopathy caused by mutations in the HBB gene that reduce or abolish β-globin chain synthesis. Because carriers are clinically silent, undetected carrier couples remain the principal route through which transfusion-dependent thalassemia is transmitted to the next generation. Confirmatory laboratory procedures such as hemoglobin electrophoresis and high-performance liquid chromatography are accurate but costly, slow and unavailable in many low-resource settings, which motivates automated screening from routinely collected data. This study presents a fusion-based decision model that identifies β-thalassemia carriers by combining two complementary modalities. A support vector machine was trained on 2,001 complete blood count reports obtained from the Punjab Thalassemia Prevention Program, and a convolutional neural network was trained on 80 peripheral blood smear images from which 64 color, texture and morphological descriptors were extracted per segmented erythrocyte. The two feature spaces were then combined through a fuzzy-logic feature-level fusion layer. All models were evaluated using a 70:30 training-validation partition. The support vector machine and the convolutional neural network attained validation accuracies of 95.00% and 95.83% respectively, whereas the proposed fuzzy fusion model reached 96.15% accuracy with a 3.85% miss-rate, 95.45% precision and 96.71% recall. The findings indicate that fusing hematological indices with erythrocyte morphology yields more dependable carrier identification than either modality used alone.