Related Experiment Video
Updated: Jun 27, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
DPCrossU-Net: a dual-branch parallel CNN-Transformer network for lung nodule segmentation
Xiya Guan1, Wen Zhu2, Fangxiang Wu3
1School of Mathematics and Statistics, Hainan Normal University, Haikou, China.
Frontiers in Oncology
|June 26, 2026
Summary
Accurate lung nodule segmentation is improved with DPCrossU-Net, a novel dual-branch network. This method effectively combines convolutional neural networks (CNNs) and Vision Transformers (ViTs) for enhanced early lung cancer detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Accurate lung nodule segmentation in CT images is critical for early lung cancer detection and diagnosis.
- Existing segmentation models struggle with small nodules, complex boundaries, and balancing local/global feature extraction.
Purpose of the Study:
- To develop an advanced deep learning model for precise lung nodule segmentation.
- To improve the accuracy and robustness of lung nodule segmentation in CT scans.
Main Methods:
- Proposed DPCrossU-Net, a dual-branch parallel encoder-decoder network integrating CNN and Vision Transformer (ViT) features.
- Employed a Cross-Attentive Fusion (CAF) module for adaptive combination of local texture and global semantic information.
- Incorporated multi-scale atrous convolutions and a dual-branch Detail Context Fusion (DCF) block for enhanced small nodule sensitivity and boundary reconstruction.
Main Results:
- DPCrossU-Net achieved a Dice score of 85.89% on the LIDC-IDRI dataset.
- Outperformed baseline U-Net, demonstrating superior performance in segmenting small nodules and complex cases.
- Showcased effectiveness in handling challenging segmentation scenarios with intricate boundaries and varied nodule sizes.
Conclusions:
- Synergistic combination of parallel CNN-Transformer feature extraction and adaptive fusion significantly enhances lung nodule segmentation accuracy.
- DPCrossU-Net offers a robust and clinically applicable solution for improved early lung cancer analysis.
- The model holds potential for supporting future intelligent diagnostic systems in radiology.
