Image Classification with Consistent Supporting Evidence
This work addresses the need for trustworthy AI in healthcare by improving model interpretability, though it is incremental as it builds on existing evidence-based methods.
The paper tackled the problem of building trust in healthcare ML models by ensuring that the supportive evidence for predictions is both compatible and sufficient, proposing measures and regularizers to promote consistency. They demonstrated this approach in edema severity grading from chest radiographs, achieving competitive performance while supporting interpretation.
Adoption of machine learning models in healthcare requires end users' trust in the system. Models that provide additional supportive evidence for their predictions promise to facilitate adoption. We define consistent evidence to be both compatible and sufficient with respect to model predictions. We propose measures of model inconsistency and regularizers that promote more consistent evidence. We demonstrate our ideas in the context of edema severity grading from chest radiographs. We demonstrate empirically that consistent models provide competitive performance while supporting interpretation.