SV-UNet: Attention-based Fully Convolutional Network with Transfer Learning for Multimodal Infarct Segmentation

DOI

10.2991/978-94-6463-540-9_60How to use a DOI?

Keywords

Ischemic Stroke; MRI Segmentation; Squeeze and Excitation Network; Transfer Learning; Fully Convolutional Network

Abstract

Ischemic stroke has a devastating impact on global health, causing both death and disability. Automatic, accurate segmentation of these stoke areas, or infarctions, from Magnetic Resonance Imaging (MRI), can aid clinicians in personalized therapeutic strategies. Recent advances in merging fully convolutional networks with transfer learning show a promising outlook, but they rarely focus on multi-modalities analysis and leverage channel-wise anatomical information to improve segmented performance. The research introduces an attention-based SV-UNet model designed to identify infarctions in two MRI modalities: Diffusion-Weighted Imaging (DWI) and T1-Weighted (T1w) images. This model derives from the UNet architecture as the backbone, employing a pre-trained VGG16 model as a shared encoder connecting to two decoders with identical architecture. In each up-convolution operation, a Squeeze-and-Excitation Network is integrated to enhance feature restoration by analyzing global information. For comparison, a VGG16-Dual-UNet is established as the benchmark. This architecture is identical to the SV-UNet, except for the removal of the SENet module. The research evaluates the two networks using two datasets: Anatomical Tracings of Lesions After Stroke 2.0R and Ischemic Stroke Lesion Segmentation 2022. The study demonstrates that SV-UNet outperforms the baseline model in detecting small stroke lesions (minority pixels) within DWI data. While performance on T1w data remains comparable, the superior sensitivity in DWI data suggests promise for improved clinical applications.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Han Xu
PY  - 2024
DA  - 2024/10/16
TI  - SV-UNet: Attention-based Fully Convolutional Network with Transfer Learning for Multimodal Infarct Segmentation
BT  - Proceedings of the 2024 2nd International Conference on Image, Algorithms and Artificial Intelligence (ICIAAI 2024)
PB  - Atlantis Press
SP  - 597
EP  - 610
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-540-9_60
DO  - 10.2991/978-94-6463-540-9_60
ID  - Xu2024
ER  -