Dual Encoder Cross-Shape Transformer Network for Medical Image Segmentation in Internet of Medical Things for Consumer Health

In emerging consumer healthcare, high-performance and robust medical image segmentation methods are essential for personalized diagnosis and treatment. Thus, early screening of aneurysms reduces the risk of aneurysm rupture and provides therapeutic and preventive measures. However, uncontrollable fa...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on consumer electronics 2025-01, p.1-1
Main Authors: Zhao, Yawu, Wang, Shudong, Zhang, Yulin, Ren, Yande, Zhang, Yuanyuan, Pang, Shanchen
Format: Article
Language:English
Subjects:
Aneurysm
Aneurysm image
Computational modeling
Consumer electronics
Convolutional neural networks
cross-shaped attention
Feature extraction
Image segmentation
internet of medical thing
Medical diagnostic imaging
semantic segmentation
Semantics
Shape
transformer
Transformers
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In emerging consumer healthcare, high-performance and robust medical image segmentation methods are essential for personalized diagnosis and treatment. Thus, early screening of aneurysms reduces the risk of aneurysm rupture and provides therapeutic and preventive measures. However, uncontrollable factors such as uncertainty in the size and location shape of tumors in medical images a significant challenge to medical image segmentation. These factors make extracting high-quality features from aneurysm images difficult, resulting in poor segmentation. Then, we designed a dual encoder cross-shape transform network (DECSTNet) to capture aneurysm feature information. The dual encoder structure can extract aneurysm feature information at different scales, the adaptive dynamic feature fusion module can fuse features at different scales between the encoders, and the cross-shape window transform layer can compute the width and height of the image in parallel for local self-attention, which enhances the interactive capability of the telematic information while realizing the complementarity of the local information. Through extensive experiments, we demonstrate the excellent segmentation performance of DECSTNet on the private and three public datasets. Noteworthy3our method has significant segmentation performance and fewer parameters, making it well-suited to be deployed on IoMT diagnostic platforms for medical image segmentation to promote healthy patient consumption.
ISSN:0098-3063
1558-4127
DOI:10.1109/TCE.2025.3526801