SFB-net for cardiac segmentation: Bridging the semantic gap with attention

TL;DR

This paper introduces SFB-net, a novel cardiac segmentation model combining convolutional and swin transformer layers to effectively model long-range dependencies and improve segmentation accuracy across datasets.

Contribution

The paper presents SFB-net, a new architecture that bridges the semantic gap in cardiac segmentation by integrating spatial attention and high-level semantic features.

Findings

Achieved 92.4% Dice score on ACDC dataset, outperforming previous methods.

Obtained 87.99% Dice score on M&M's dataset, demonstrating good generalization.

Effectively models long-range dependencies in cardiac image segmentation.

Abstract

In the past few years, deep learning algorithms have been widely used for cardiac image segmentation. However, most of these architectures rely on convolutions that hardly model long-range dependencies, limiting their ability to extract contextual information. In order to tackle this issue, this article introduces the Swin Filtering Block network (SFB-net) which takes advantage of both conventional and swin transformer layers. The former are used to introduce spatial attention at the bottom of the network, while the latter are applied to focus on high level semantically rich features between the encoder and decoder. An average Dice score of 92.4 was achieved on the ACDC dataset. To the best of our knowledge, this result outperforms any other work on this dataset. The average Dice score of 87.99 obtained on the M\&M's dataset demonstrates that the proposed method generalizes well to data from different vendors and centres.