An AI-Powered Clinical Decision Support System for Leukemia Type and Stage Detection Using Hybrid Machine Learning and Deep Learning Models

Artificial intelligence in leukemia diagnosis has attracted considerable interest for enhancing diagnostic precision and streamlining clinical workflows in hematopathology. Although deep learning has progressed in image-based classification and detection, existing systems typically rely on unimodal inputs either structured clinical data or microscopic blood smears resulting in compromised robustness, interpretability, and practical utility. The proposed work presents a hybrid multi-stage prototype integrating clinical parameters with image analysis to mirror real-world diagnostics. In the first stage, a Support Vector Machine (SVM) classifies leukemia presence and subtypes (AML, ALL, CML, CLL, or Normal) using complete blood count and bone marrow metrics, leveraging SVM's strength in nonlinear biomedical pattern recognition with limited samples. Leukemia-positive cases trigger a second stage featuring Swin Transformer for hierarchical feature extraction and Graph Neural Network (GNN) for spatial relationship modeling in white blood cell images, determining maturation stages (Progenitor, Precursor, Early). Reliability enhancements include class-balancing, confidence-based prediction rejection, and a Streamlit clinical interface, yielding superior performance over conventional single-stage approaches.