Back

Leakage-Resistant Models for MCI-to-Dementia Conversion Risk Prediction

Balbino TchoutzineJuly 24, 2026Ongoing

Context

This pre-engineering research internship is part of the first semester of my final year (ING-5) at ENSPY, supervised by Pr Tsopze Norbert (University of Yaoundé I), in collaboration with Pr Bernard Fongang (University of Texas, USA).

Subject

Building leakage-resistant models for predicting the risk of conversion from Mild Cognitive Impairment (MCI) to dementia, over a 2 to 5 year horizon, across three reference cohorts in neuroscience research:

  • NACC (National Alzheimer's Coordinating Center)
  • ADNI (Alzheimer's Disease Neuroimaging Initiative)
  • Framingham (Framingham Heart Study)

The data combines clinical information, imaging, and biomarkers.

Approach

The work focuses on three areas:

  • External validation: assessing how well models generalize across cohorts
  • Calibration: ensuring predicted risk scores reflect real-world probabilities
  • Interpretable baselines: reference models (Cox regression, XGBoost) analyzed with SHAP to understand the identified risk factors

Status

This project is ongoing. Results and a more detailed methodology description will be added as the internship progresses.