International testing and refinement of AI algorithms predicting acute leukemia subtypes from routine laboratory data - PubMed
7 hours ago
- #medical research
- #leukemia diagnosis
- #AI in healthcare
- International testing of AI algorithms for predicting acute leukemia subtypes using routine lab data.
- Diverse cohort of 6206 leukemia patients from 20 centers used to test AI tool.
- Initial AUROC metrics: 0.94 for AML, 0.98 for promyelocytic subtype, 0.84 for ALL.
- Confidence cutoff predictions excluded 70.8-92.5% of patients.
- Improved accuracy via ensemble of Isolation Forest and Local Outlier Factor, increasing AML AUROC to 0.84.
- Retrained algorithm for pediatric patients.
- Multiple authors disclosed consultancy roles, advisory roles, and research funding from pharmaceutical companies.