Machine Learning Explainability in Breast Cancer Survival
Problem
Black-box ML models that predict breast cancer survival are hard for clinicians to trust, hindering clinical adoption.
Why it matters
Explainable survival predictions could improve treatment decisions and acceptance of ML in cancer care.
What the AI did
An ML model predicted 10-year overall survival from Netherlands Cancer Registry data, and LIME and SHAP were used to interpret its predictions.
Results
LIME and SHAP found largely consistent in explaining feature contributions; mismatches reveal 'turning points' between survival outcomes
Impact
The study showed explainability techniques can support acceptance of ML in clinical settings, though translation to real-life scenarios remains open.
Record
| Venue | Studies in health technology and informatics |
|---|---|
| Published | 2020 |
| Work type | book-chapter |
| Authors | Tom Jansen, Gijs Geleijnse, Van Maaren Marissa, Mathijs P. Hendriks, Annette ten Teije, Arturo Moncada‐Torres |
| Citations | 38 (OpenAlex, at import time) |
| References | 2 |
| Model or method | LIME and SHAP (base ML model not stated in source) |
| Organization | Netherlands Cancer Registry (data source) |
| Primary source | https://doi.org/10.3233/shti200172 |
| Open access | hybrid |
| Keywords | machine learning explainability, LIME, SHAP, breast cancer, survival prediction |
Importance rating
significant — thousands to millions affected
Breast cancer survival prediction affects large patient populations; explainability for clinical trust addresses a genuine barrier to ML adoption in cancer care, though abstract states no scale.