Genetically Aligned Patient Representations Improve Hematological Diagnosis

Researchers developed a framework that aligns single-cell white blood cell images with genetic data (karyotypes and mutations) to improve hematological cancer diagnosis. Using a two-stage training approach combining self-supervised vision learning and supervised contrastive alignment, the model outperforms existing histopathology foundation models and enables disease retrieval based on genetic alterations.

This study represents a meaningful advance in multimodal medical AI by addressing a specific clinical workflow challenge in hematology. Blood cancer diagnosis inherently requires integration of morphological, cytogenetic, and molecular data—a complexity that standard vision-only models cannot capture. The researchers' two-stage approach first builds robust visual representations through self-supervised learning on 1500+ patients, then grounds these representations in genetic reality through contrastive alignment on acute myeloid leukemia cases.

The broader context reflects a growing recognition that medical AI performs better when trained on clinically integrated data rather than isolated modalities. Recent successes in aligning histopathology with genomics have established this principle, but hematology presents unique opportunities given the discrete nature of single cells and the prevalence of genetic testing in clinical protocols. This work extends those successes into a specialized domain where the alignment is particularly natural and valuable.

From a healthcare perspective, this framework could accelerate diagnostic workflows by automating pattern recognition across multiple data types simultaneously, reducing clinician cognitive load and potential for missed correlations between morphology and genetics. The availability of open-source code and model weights enables broader adoption and validation across institutions. The retrieval capabilities offer additional clinical utility for rare disease recognition and mutation pattern discovery.

Future developments will likely focus on expanding the framework beyond AML to other hematological malignancies, validating performance on prospective patient cohorts, and integrating additional modalities like flow cytometry data. The work establishes a template for hematology-specific AI that respects clinical realities rather than forcing generic architectures onto specialized diagnostic problems.

• →Genetically-aligned patient encoders significantly improve hematological diagnostic accuracy compared to slide-level histopathology models alone.
• →The two-stage training strategy combines self-supervised vision learning with supervised genetic alignment to ground visual representations in molecular reality.
• →The framework enables off-the-shelf retrieval capabilities for diseases and genetic alterations, supporting clinical discovery workflows.
• →Integration of genetic data into patient representations creates AI systems that align with existing hematological diagnostic protocols.
• →Open-source release of code and weights facilitates adoption and validation across multiple healthcare institutions.