GBIR: A Novel Gaussian Iterative Method for Medical Image Reconstruction

ICLR 2025 Conference Submission1444 Authors

Keywords: Medical Image Reconstruction, Computed Tomography, Nuclear Magnetic Resonance Imaging

Abstract:

Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) are crucial diagnostic tools, but undersampling techniques like Sparse-View CT (SV-CT) and Compressed-Sensing MRI (CS-MRI), aimed at reducing patient exposure and scan time, make image reconstruction more challenging. While deep learning-based reconstruction (DLR) methods have made significant strides, they face limitations in adapting to varying scan geometries and handling diverse patient data, hindering widespread clinical use.
In this paper, we propose a novel **G**aussian-**B**ased **I**terative **R**econstruction (**GBIR**) framework that uses learnable Gaussians representations for personalized medical image reconstruction, addressing the shortcomings of DLR methods. GBIR optimizes case-specific parameters in an end-to-end fashion, enabling better generalization and flexibility under sparse measurements. Additionally, we introduce the **M**ulti-**O**rgan Medical Image **RE**construction (**MORE**) dataset, comprising over 70,000 CT and MRI slices across multiple body parts and conditions.
Our experiments show that GBIR outperforms state-of-the-art methods in both accuracy and speed, offering a robust solution for personalized medical image reconstruction.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1444