LeukoAlert-project：Transforming routine blood counts into a leukemia detection system with interpretable AI
Description
Title of Study: LeukoAlert: An Interpretable AI Framework for Large-Scale Leukemia Screening Using Routine Complete Blood Count (CBC) Data Overview: This repository contains the source code and de-identified datasets supporting the findings of the "LeukoAlert" study. We present an interpretable artificial intelligence framework designed to screen for leukemia and its subtypes using routine Complete Blood Count (CBC) data. Repository Contents: Dataset: The dataset consists of de-identified clinical records provided in standardized CSV format. To ensure rigorous evaluation, the data are stratified into three distinct cohorts: Training Cohort: Over 20,000 records used for model development and hyperparameter optimization (utilizing internal 5-fold cross-validation). Validation Cohort (Multi-center): Over 300,000 records from independent centers (Sites A-G), used to assess model generalization. Real-world Evaluation Cohort: Over 50,000 records representing real-world clinical application scenarios. Features: Each record includes 72 CBC parameter.
Files
Institutions
- Shanghai Jiao Tong UniversityShanghai