U-Net based on multi-head attention mechanism

Guangyuan Zhang; Yingxiang Lu; Tingzhi Qiu; Wei Li

doi:10.1117/12.3012314

21 December 2023 U-Net based on multi-head attention mechanism

Guangyuan Zhang, Yingxiang Lu, Tingzhi Qiu, Wei Li

Proceedings Volume 12970, Fourth International Conference on Signal Processing and Computer Science (SPCS 2023); 1297043 (2023) https://doi.org/10.1117/12.3012314
Event: Fourth International Conference on Signal Processing and Computer Science (SPCS 2023), 2023, Guilin, China

Abstract

Detection of anatomical landmarks in medical images plays a crucial role in understanding anatomy and facilitating automated processing. In recent years, various deep neural network methods have been developed for automated landmark detection. However, deep neural networks suffer from limitations in expressive power and are prone to overfitting. In this study, we propose a novel approach that combines the multi-head attention mechanism with U-Net architecture to enhance the expressive capability of deep learning. The proposed model consists of encoding and decoding modules. The encoding module leverages a dual multi-head attention mechanism to learn local features, while the decoding module employs a depth-wise separable convolutional sequence corresponding to the encoding module. These two modules are concatenated through skip connections. We evaluate our model on an open-source dataset of lateral skull x-ray images, which includes 400 images with 19 landmark points in each image. Notably, experimental results demonstrate that our proposed model outperforms known open-source models in terms of performance, providing evidence for the effectiveness of our proposed approach.

(2023) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Guangyuan Zhang, Yingxiang Lu, Tingzhi Qiu, and Wei Li "U-Net based on multi-head attention mechanism", Proc. SPIE 12970, Fourth International Conference on Signal Processing and Computer Science (SPCS 2023), 1297043 (21 December 2023); https://doi.org/10.1117/12.3012314

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available