Deep Learning-assisted 3D Segmentation for Monitoring Cartilage Regeneration in Knee MRI Scans
Collaboration
R. Aabling and M. Pedersen, Department of Clinical Medicine, Aarhus University
Research Background
Osteoarthritis leads to progressive degeneration of knee joint cartilage. Recent research suggests that stem cell therapy can support cartilage regeneration. This study evaluates the efficacy of intra-articular stem cell injections using longitudinal MRI scans to monitor morphological changes in femoral and tibial cartilage.
Method
MRI scans acquired at baseline (pre-treatment), mid-treatment, and study completion are analyzed using a dedicated segmentation pipeline built on a modified multi-planar U-Net architecture (Perslevet al., 2019) in Python. The pipeline involves:
Pretraining: Pretrain a modified Multiplanar U-Net on large, annotated dataset of knee MRI scans with a similar MRI protocol
Transfer Learning: Finetune the model on one fully annotated scan for cartilage segmentation
Iterative refinement:
a) Predict segmentation on a new patient scan
b) Manual, expert correction of the segmentation
c) Retrain model with the new sample