AUTOMATIC CLASSIFICATION OF LEUKEMIA CELLS VIA FLOW CYTOMETRY AND MITRAL REGURGITATION DETECTION USING MACHINE LEARNING ALGORITHMS

Abstract

Early and accurate diagnosis of hematological malignancies and cardiac valvular disorders remains critical for effective treatment planning and improved patient outcomes. This research presents dual machine learning frameworks addressing automated leukemia cell classification from flow cytometry data and mitral regurgitation detection from echocardiographic imaging. For leukemia classification, we developed ensemble learning models processing multi-parameter flow cytometry measurements to distinguish acute lymphoblastic leukemia, acute myeloid leukemia, and normal cell populations with 96.3% accuracy. The framework employs feature engineering extracting immunophenotypic signatures, dimensionality reduction via UMAP, and gradient boosting classifiers handling class imbalance inherent in clinical datasets. For mitral regurgitation detection, we implemented convolutional neural networks analyzing Doppler echocardiography images, achieving 94.7% sensitivity and 92.8% specificity in identifying pathological regurgitation. The integrated system incorporates uncertainty quantification providing confidence scores for clinical decision support, explainability mechanisms highlighting discriminative features, and data augmentation strategies addressing limited annotated medical datasets. Validation on multicenter clinical cohorts totaling 3,847 leukemia samples and 2,156 echocardiographic studies demonstrates robust generalization across diverse patient populations and imaging equipment. The combined framework reduces diagnostic time by 73% compared to manual analysis while maintaining diagnostic accuracy comparable to expert hematologists and cardiologists, enabling faster treatment initiation and potentially improving survival rates in time-sensitive conditions.