RESEARCH / XAI / MEDICAL IMAGING
EXPLAINABLE AI FOR MEDICAL IMAGING.
Research that does not stop at model predictions; it also examines where a model looked to make them.
BACKGROUND
In medical imaging AI, performance metrics alone do not fully explain a model's decisions. This work connects explainability to implementable evaluation.
RESEARCH QUESTION
- 01Which regions does the model use as evidence?
- 02How do explanation regions differ by method?
- 03How much does generation time vary?
- 04Are explanations stable under input changes?
METHODS
SHAPLIMEGRAD-CAM
EXPERIMENT
- — Dataset: chest X-ray images
- — Model: DenseNet121
- — Preprocessing: normalization and resizing
- — Evaluation: classification performance, explanation region, generation time
LIMITATIONS
- 01
Constraints in case count and data distribution
- 02
Medical evaluation needs expert review
- 03
Evaluation criteria for explanation methods still need refinement