Related Experiment Video
Updated: Sep 20, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.6K
DECE-Net: a dual-path encoder network with contour enhancement for pneumonia lesion segmentation
Tianyang Wang1, Xiumei Li1, Ruyu Liu1
1Hangzhou Normal University, School of Information Science and Technology, Hangzhou, China.
Journal of Medical Imaging (Bellingham, Wash.)
|May 26, 2025
Summary
A novel deep learning model, DECE-Net, accurately segments pneumonia lesions in CT scans, overcoming challenges like low contrast and variable shapes. This advancement aids early detection and treatment for better patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Early-stage pneumonia detection is challenging due to difficulties in segmenting lesions in CT images.
- Low contrast, and variations in lesion size and shape hinder accurate segmentation, often leading to missed treatment windows.
Purpose of the Study:
- To propose DECE-Net, an automated segmentation network for precise pneumonia lesion identification in CT images.
- To address the limitations of existing methods in segmenting pneumonia lesions with low contrast and diverse characteristics.
Main Methods:
- DECE-Net enhances the U-Net architecture with an additional encoder path for contour feature extraction.
- It utilizes attention multi-scale feature fusion and a feature fusion attention connection module to integrate multi-level features.
- Multi-point deep supervision is applied across different scales to refine segmentation accuracy.
Main Results:
- DECE-Net achieved high mean Intersection over Union (mIoU) scores on four public COVID-19 datasets: 80.76%, 84.59%, 84.41%, and 78.55%.
- The model demonstrated state-of-the-art performance in segmenting small pneumonia lesion areas.
Conclusions:
- The proposed DECE-Net effectively overcomes segmentation challenges in CT images for pneumonia detection.
- The network shows significant promise for improving early diagnosis and treatment of pneumonia, particularly for subtle lesions.
