Carrie L. Langstraat

Pretrained vision transformers can achieve state-of-the-art medical image segmentation with significantly less labeled data, outperforming fully supervised convolutional networks even when trained on only 25% of the dataset.