Related Experiment Video
Updated: Jan 10, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
A progressive fusion network for endoscopic medical image segmentation.
Lihong Fu1,2, Zhengping Li3,4, Chao Xu1,2
1School of Integrated Circuits, Anhui University, HeFei, 230601, China.
Scientific Reports
|November 29, 2025
Summary
A new progressive fusion network (PFNet) enhances endoscopic image segmentation by effectively combining local details and global context. This method improves diagnostic accuracy for complex organ and tissue identification.
Area of Science:
- Medical image analysis
- Computer vision
- Artificial intelligence in healthcare
Background:
- Accurate endoscopic image segmentation is crucial for medical diagnosis and procedural guidance.
- Existing segmentation methods struggle with complex morphologies, fuzzy boundaries, and similar textures due to limitations in integrating local and global information.
Purpose of the Study:
- To develop an advanced endoscopic image segmentation method that effectively utilizes both local details and global semantic information.
- To improve the accuracy and efficiency of organ and tissue segmentation in endoscopic imaging.
Main Methods:
- Proposed a Progressive Fusion Network (PFNet) utilizing a Pvtv2 with Transformer backbone for multi-scale global feature extraction.
- Introduced a Noise Filtering Attention Module (NFAM) to enhance feature semantics and suppress noise.
- Developed a Boundary and Location Awareness Module (BLAM) and Auxiliary Information Embedding Module (AIEM) to integrate boundary and positional information.
- Employed a Feature Fusion Module (FFM) for iterative layer-by-layer fusion to recover global semantics and local details.
Main Results:
- PFNet demonstrated superior performance over state-of-the-art methods across multiple endoscopic datasets (Ureter, Re-TMRS, Kvasir, CVC-ClinicDB, CVC-ColonDB, ETIS, CVC-300).
- Achieved high mDice scores, including 91.07% on the Re-TMRS dataset and 93.09% on the CVC-ClinicDB dataset.
Conclusions:
- The proposed PFNet effectively addresses the limitations of previous methods by synergistically leveraging local and global features for precise endoscopic image segmentation.
- PFNet offers a significant advancement in automated analysis of endoscopic images, potentially enhancing diagnostic efficiency and accuracy in clinical practice.
