Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

A Data-scalable Transformer for Medical Image Segmentation: Architecture, Model Efficiency, and Benchmark

About

Transformers have demonstrated remarkable performance in natural language processing and computer vision. However, existing vision Transformers struggle to learn from limited medical data and are unable to generalize on diverse medical image tasks. To tackle these challenges, we present MedFormer, a data-scalable Transformer designed for generalizable 3D medical image segmentation. Our approach incorporates three key elements: a desirable inductive bias, hierarchical modeling with linear-complexity attention, and multi-scale feature fusion that integrates spatial and semantic information globally. MedFormer can learn across tiny- to large-scale data without pre-training. Comprehensive experiments demonstrate MedFormer's potential as a versatile segmentation backbone, outperforming CNNs and vision Transformers on seven public datasets covering multiple modalities (e.g., CT and MRI) and various medical targets (e.g., healthy organs, diseased tissues, and tumors). We provide public access to our models and evaluation pipeline, offering solid baselines and unbiased comparisons to advance a wide range of downstream clinical applications.

Yunhe Gao, Mu Zhou, Di Liu, Zhennan Yan, Shaoting Zhang, Dimitris N. Metaxas• 2022

Related benchmarks

TaskDatasetResultRank
Medical Image SegmentationBUSI (test)
Dice81.4
228
Medical Image SegmentationISIC 2018--
187
Skin Lesion SegmentationISIC 2018 (test)
Dice Score88.25
143
Skin Lesion SegmentationISIC 2017 (test)
Dice Score87.23
134
Medical Image SegmentationISIC 2017--
102
Skin Lesion SegmentationISIC 2018
Dice Coefficient87.3
94
Skin Lesion SegmentationPH2
DIC0.8885
87
Skin Lesion SegmentationPH2 (test)
DSC85.85
70
2D skin lesion segmentationISIC 2017
mIoU78.35
60
Medical Image SegmentationISIC (test)
IoU0.8114
55
Showing 10 of 28 rows

Other info

Follow for update