CK-ATTnet: Medical image segmentation network based on convolutional kernel attention.

Autor: Cai B; College of Computer Science and Cyber Security, Chengdu University of Technology, Chengdu 610059, China; College of Industrial Technology, Chengdu University of Technology, Yibin 644000, China. Electronic address: caibiao@cdut.edu.cn., Liu M; College of Computer Science and Cyber Security, Chengdu University of Technology, Chengdu 610059, China. Electronic address: 903442845@qq.com., Lu Z; College of Computer Science and Cyber Security, Chengdu University of Technology, Chengdu 610059, China. Electronic address: tyutlulu@163.com., Liu M; School of Data Science and Artifical Intelligence, Wenzhou University of Technology, Wenzhou 325035, China. Electronic address: liumz@cdut.edu.cn.
Jazyk: angličtina
Zdroj: Computers in biology and medicine [Comput Biol Med] 2024 Dec; Vol. 183, pp. 109300. Date of Electronic Publication: 2024 Oct 28.
DOI: 10.1016/j.compbiomed.2024.109300
Abstrakt: The medical image partition model has a wide range of application prospects in medical diagnosis and treatment and has become an important auxiliary method to improve the diagnostic level by medical imaging analysis. After the feature extraction ability of the convolutional neural network (CNN) reached a bottleneck, the form of feature extraction represented by Transformer has made significant achievements in the medical image domain in recent years. However, the structure of Transformer is relatively fixed, the cost of computer resources is large, and it is difficult to adjust the model structure according to the complex medical imaging segmentation task. To better adapt to the limitation of clinical diagnostic equipment on the parameter scale of the network model, this paper proposed a CK-ATTnet based on the convolutional kernel attention mechanism. CK-ATTnet uses the depthwise separable convolution attention mechanism, which completely innovates the way that the attention mechanism is used on the original image in the traditional model. In addition, CK-ATTnet realizes the new normal form of applying the attention mechanism to the convolutional kernel for feature extraction for the first time. This design further improves the local feature acquisition ability of the convolutional kernel and does not require additional hardware enhancement requirements after applying the attention mechanism to the model. Compared with other CNN models, CK-ATTnet can extract more accurate and fine-grained features; compared with Transformer-based models, it has fewer learning parameters. Experimental results show that CK-ATTnet exhibits better segmentation performance and fewer learning parameters than other models in multiple datasets and has very good application prospects.
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2024 Elsevier Ltd. All rights reserved.)
Databáze: MEDLINE
Externí odkaz: