Spinal Cord Segmentation Generalizable Across Datasets
My idea is to train a deep learning model on multiple (4) spinal cord segmentation datasets to improve generalizability to new contrasts, vendors, pathologies, etc… My project aims to train the nnU-Net model architecture, a state-of-the-art deep learning architecture for biomedical segmentation, on four aggregated datasets and compare its generalizability capabilities with the four specific models trained on each individual dataset. I will conclude by comparing the two approaches on a fifth and sixth dataset outside of the training domain.