Related Experiment Video
Updated: Dec 15, 2025
07:03
Medical-grade Sterilizable Target for Fluid-immersed Fetoscope Optical Distortion Calibration
Published on: February 23, 2017
8.0K
Fast geometric distortion correction using a deep neural network: Implementation for the 1 Tesla MRI-Linac system.
Mao Li1, Shanshan Shan1, Shekhar S Chandra1
1School of Information Technology and Electrical Engineering, University of Queensland, Brisbane, QLD, 4067, Australia.
Medical Physics
|July 11, 2020
Summary
A novel deep neural network corrects gradient nonlinearity distortions in 1 Tesla MRI-Linac systems, enhancing accuracy for real-time image-guided radiotherapy and improving patient safety.
Area of Science:
- Medical Physics
- Radiotherapy Technology
- Artificial Intelligence in Medicine
Background:
- Magnetic Resonance Imaging-Linear Accelerator (MRI-Linac) systems enable real-time monitoring during radiotherapy.
- Gradient nonlinearity (GNL)-induced distortions in MRI hinder precise image-guided radiotherapy due to inaccurate geometric and anatomical representation.
- Accurate geometric correction is crucial for effective tumor targeting and minimizing healthy tissue exposure.
Purpose of the Study:
- To develop and validate a deep neural network for correcting GNL-induced geometric distortions in 1 Tesla (T) MRI-Linac systems.
- To enable accurate anatomical and geometric representation for improved radiotherapy delivery.
- To enhance the feasibility of routine implementation of MRI-Linac systems in clinical practice.
Main Methods:
- A deep fully connected neural network was designed to learn the relationship between distorted and undistorted MR image spaces.
- A comprehensive dataset from phantom measurements, covering regions inside and outside the spherical volume, was used for network training.
- The network was trained to characterize subtle deviations of the GNL field within the entire region of interest (ROI).
Main Results:
- The proposed deep neural network successfully corrected severe geometric distortions across the entire ROI in MR images from a 3D phantom and a human volunteer.
- The correction accuracy achieved an error less than the pixel size, demonstrating high precision.
- The network exhibited significant computational efficiency improvements compared to existing methods.
Conclusions:
- The study demonstrated the feasibility of using a deep neural network to characterize GNL field deviations in 1T MRI-Linac systems.
- The developed method shows strong potential for routine implementation in real-time MRI-guided radiotherapy.
- This advancement promises to improve the accuracy and safety of radiotherapy treatments.
