Using the medicaltorch library to perform semantic segmentation on Medical Decathlon datasets.
Used a U-Net with the Dice coefficient as the loss function, the Adam optimizer and a learning rate scheduler.
Results on the Heart dataset:
Sample prediction on a 2D slice post training on the Heart dataset:
