Related Experiment Video
Updated: Jun 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Implicit neural representation for medical image reconstruction
Yanjie Zhu1, Yuanyuan Liu1, Yihang Zhang1,2
1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, People's Republic of China.
Physics in Medicine and Biology
|June 2, 2025
Summary
Implicit neural representation (INR) offers a novel solution for medical image reconstruction by continuously modeling signals, overcoming limitations of traditional and deep learning methods that require extensive data. This approach enhances image quality and detail capture.
Area of Science:
- Medical Imaging
- Computational Imaging
- Artificial Intelligence
Background:
- Medical image reconstruction is an ill-posed inverse problem requiring high-quality images from limited data.
- Traditional methods use regularization terms, while deep learning requires large datasets, which are scarce in medical imaging.
- Implicit Neural Representation (INR) provides a continuous, flexible image representation using spatial coordinates.
Purpose of the Study:
- To review Implicit Neural Representation (INR) techniques for medical image reconstruction.
- To highlight the growing impact and benefits of INR in medical imaging.
- To discuss the advantages, limitations, and future directions of INR in this field.
Main Methods:
- Reviewing existing literature on INR-based medical image reconstruction.
- Analyzing the application of INR in both image and measurement domains.
- Evaluating the effectiveness of INR in capturing fine details and complex structures.
Main Results:
- INR offers a flexible and continuous representation of medical images.
- INR effectively captures fine details and complex structures compared to discrete methods.
- INR shows promise in addressing the challenges of data scarcity in medical image reconstruction.
Conclusions:
- Implicit Neural Representation (INR) is a powerful emerging technique for medical image reconstruction.
- INR presents significant advantages over traditional and supervised deep learning methods, particularly regarding data requirements.
- Further research into INR holds potential for advancing medical imaging quality and applications.
