HealthML

HealthML

A major challenge in clinical AI is building models that generalize across institutions. Models trained on one hospital’s data often fail when applied elsewhere —known as the Model Export Problem.

This project develops a clinical data infrastructure using MIMIC-IV on Google Cloud Platform, with an interactive dashboard covering seven disease categories. The core research focuses on dementia: we build a dementia classification model trained on MIMIC-IV and evaluate its exportability to the ADNI (Alzheimer’s Disease Neuroimaging Initiative) dataset, investigating why model performance degrades across datasets and how generalizability can be improved.


Targeted Conferences