AI Diagnostic Models for Blood Disorders Lack Validation
Current research models have yet to enter interventional clinical trials for patients with myeloproliferative neoplasms.
Updated on Oct. 31, 2026 in Artificial Intelligence
Live Poll
Should medical AI tools require clinical validation before being used to inform patient care?
A review of current research finds that while machine learning models show promise in diagnosing and grading myeloproliferative neoplasms, none have reached the stage of interventional clinical testing. These AI tools are currently confined to research cohorts and internal validation studies.
Why it matters
Bridging the gap between technical benchmarks and demonstrated clinical utility remains a hurdle for hematological AI. Without prospective validation, these models cannot yet be used to guide treatment decisions for myelofibrosis patients.
The FIBOM-AI model uses 27 complete blood count variables to detect fibrosis, demonstrating an AUC of 0.92 in external testing and 85.2% accuracy in prospective practice. Other foundation models for disease subtyping report AUCs of 0.89 to 0.94.
The players
Blood Reviews
A scientific journal specializing in hematology research and clinical updates.
The details
Researchers compiled results from studies using digital pathology and blood count data to classify neoplasms and fibrosis grades. Digital pathology models use automated image analysis to assess bone marrow, while prognostic models like AIPSS-MF integrate clinical variables to estimate survival. These tools are currently research-stage and rely on retrospective data rather than live clinical trials.
Timeline
August 2026: Investigators completed their systematic literature search.
October 23, 2026: Findings were published in Blood Reviews.
The Tech Race
The field is shifting from simple proof-of-concept models to more complex diagnostic tools like the FIBOM-AI platform. However, it currently trails behind standard hematological scores in terms of regulatory and clinical integration.
For hematologists, these tools are currently limited to research and auxiliary analysis rather than primary diagnostic care. No patients should expect these AI models to dictate their treatment plans until prospective multicenter trials are conducted.
The takeaway
While AI models are demonstrating high accuracy in research settings, they remain unsupported by the clinical validation required for patient care. The field is now looking for head-to-head comparisons against standard diagnostic scores to determine real-world performance.
Further reading
Learn more about the latest developments in Artificial Intelligence.
Source note: This article includes information reported by Ajmc.
Live Poll
Should medical AI tools require clinical validation before being used to inform patient care?
