Extraction of Novel Features and Diagnosis Prediction in Myelodysplastic Neoplasm Using a Weakly Supervised Artificial Intelligence Model Based on Normal Megakaryocytes.

This study developed an AI model that can accurately classify myelodysplastic neoplasm (MDS) versus normal cases using morphological features of normal megakaryocytes in bone marrow biopsies. The model integrates deep learning components for detecting bone marrow regions and identifying megakaryocytes, along with an XGBoost-based classifier to differentiate between normal, MDS, and immune thrombocytopenic purpura cases. The model achieved exceptional accuracy, with an AUC of 0.879 for MDS versus normal classification. Feature analysis revealed that the percentage of megakaryocytes and their spatial distribution were significantly correlated with disease prediction, providing novel insights into MDS pathology. This pioneering work demonstrates the potential of AI-driven analysis of bone marrow biopsies for diagnostic assistance and uncovering new disease biomarkers.