Related Experiment Video
Updated: Aug 5, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
R3Net: Recursive Residual Refinement Network Architecture for Decoder-Free Medical Image Segmentation
Jing Huang1, Yongkang Zhao1, Yuhan Li1
1School of Computer Science and Artificial Intelligence, Wuhan University of Technology Wuhan China.
Precision Radiation Oncology
|August 1, 2026
Summary
This study introduces R3Net, an efficient encoder-only medical image segmentation model. It achieves high accuracy comparable to encoder-decoder methods with fewer parameters and less computation.
Area of Science:
- Medical Imaging
- Computer Vision
- Deep Learning
Background:
- Encoder-decoder architectures are common for medical image segmentation, offering high accuracy.
- These models often demand significant computational resources and have redundant parameters.
Purpose of the Study:
- To develop an efficient, decoder-free segmentation framework for medical images.
- To maintain competitive performance by enhancing the encoding process.
Main Methods:
- Proposing R3Net, an encoder-only segmentation architecture utilizing a recursive residual refinement (R3) mechanism.
- Recursively reusing encoder stages and fusing multiscale features via residual pathways to reconstruct high-resolution features without a decoder.
Main Results:
- R3Net demonstrated segmentation performance comparable to established encoder-decoder models across cardiac MRI, abdominal CT, and thyroid ultrasound datasets.
- The proposed R3Net model significantly reduced the number of parameters and computational complexity compared to existing methods.
Conclusions:
- R3Net offers an effective decoder-free alternative for medical image segmentation.
- Competitive dense prediction is achievable through encoder-based recursive refinement.