The objective of this repository is to apply LIME and Grad-CAM to interpret the predictions of various artificial intelligence models, including:
- K-Nearest-Neighbours
- Support vector machine
- XGboost
- Convonutional Neural Network
- Random Forest
These models were trained to classify computed tomography (CT) images of the brain into three categories:
- 0: Normal
- 1: Hemorrhagic stroke (bleeding)
- 2: Ischemic stroke
As expected, models based on vectorized images provided explanations with little medical sense. In contrast, the CNN demonstrated the ability to accurately identify regions associated with strokes.