Current problems with the model

ivadomed / model_seg_ms_mp2rage

Model repository for MS lesion segmentation on MP2RAGE data from University of Basel

MIT License

2 stars 0 forks source link

Current problems with the model #48

Closed uzaymacar closed 1 year ago

uzaymacar commented 2 years ago

We are investigating model predictions and a few problems we observe are:

Some predictions go outside the spinal cord boundary.
The predictions are square like most of the time, there might be an interpolation problem.

uzaymacar commented 2 years ago

During today's meeting, we decided to manually fix the lesion GT first as a means of trying to improve the model performance. This is further documented in #13.