Trustworthy and explainable machine learning for clinical decision support
Prediction models whose probabilities are well calibrated and whose decisions can be explained to clinicians.
Background
Clinical decision support is only useful if clinicians can trust the predicted probabilities and understand the reasoning behind a decision.
Research questions
- How can prediction models give well-calibrated probabilities on clinical data?
- How can explanations be made reliable and useful to clinicians?
Objectives
- Improve and evaluate the calibration of clinical prediction models
- Evaluate explanations for clinical transparency
- Assess performance beyond the development dataset
Approach
Calibration methods and explainable AI techniques applied to clinical tabular data.
Related experience
- Calibrated general anaesthesia triage model (manuscript under review)
The full proposal is shared privately on request. Request the full proposal · All proposals