Not started yet — this is a project I’m planning to build using the RSNA knee MRI abnormality detection dataset from Kaggle.
The plan: train a CNN classifier on the scans, then build a small visualization layer on top of the results — likely a Grad-CAM style heatmap showing which regions of a scan the model weighted most heavily, alongside the usual ROC/loss curves.
One constraint I’m building around from the start: the dataset’s competition terms restrict redistributing real patient scans. Any public visualization on this site will use anonymized or synthetic examples, not raw dataset images.
Check back once there’s a trained model and real results to show.