iSegFormer: Interactive Segmentation via Transformers with Application to 3D Knee MR Images

TL;DR

iSegFormer introduces a memory-efficient transformer model combining Swin transformer and MLP decoder, achieving high performance and efficiency in interactive 3D medical image segmentation.

Contribution

The paper presents a novel transformer architecture, iSegFormer, optimized for memory efficiency and applied to 3D medical image segmentation tasks.

Findings

High computational efficiency in 3D segmentation

Effective combination of hierarchical self-attention and MLP decoding

Improved segmentation accuracy on medical images

Abstract

We propose iSegFormer, a memory-efficient transformer that combines a Swin transformer with a lightweight multilayer perceptron (MLP) decoder. With the efficient Swin transformer blocks for hierarchical self-attention and the simple MLP decoder for aggregating both local and global attention, iSegFormer learns powerful representations while achieving high computational efficiencies. Specifically, we apply iSegFormer to interactive 3D medical image segmentation.