Text this: Triple-view feature learning for medical image segmentation