AI for the assessment and discovery of morphological-molecular biomarker relationships in hematologic malignancies

Artificial intelligence (AI) is transforming morphological assessment in hematologic malignancies by enabling automated cell classification, detection of disease-specific features, and prediction of genetic alterations from blood and bone marrow smear images. Deep learning models have achieved expert-level performance in identifying malignant patterns, uncovering morphologic-molecular associations, and providing patient-level diagnoses through multiple instance learning approaches. This review summarizes recent advances in AI-driven morphological analysis across myeloid and lymphoid neoplasms, emphasizing their impact on diagnosis and disease classification. We further examine emerging evidence linking image-derived features, such as nuclear texture, cytoplasmic granularity, and spatial cell organization, to molecular and prognostic variables. In addition, we discuss enabling technologies, including virtual staining and interpretability methods, that support clinical integration. Finally, we address key challenges, including data heterogeneity, the need for prospective validation, and interdisciplinary collaboration. AI-powered, explainable morphological analysis has the potential to advance precision hematology and personalized patient management.