MRI Gibbs-ringing artifact reduction by means of machine learning using convolutional neural networks

Area of Science:

• Medical Imaging
• Machine Learning
• Image Processing

Background:

• Gibbs-ringing artifacts in MRI arise from undersampling k-space data.
• Conventional methods reduce artifacts by smoothing, often sacrificing image detail.
• A need exists for artifact reduction techniques that preserve image fidelity.

Purpose of the Study:

• To develop and validate a machine learning approach for reducing Gibbs-ringing artifacts in MRI.
• To utilize a convolutional neural network (CNN) for artifact identification and removal.
• To ensure the developed method maintains data fidelity and image quality.

Main Methods:

• A deep CNN was trained on 17,532 T2-weighted (T2W) brain images to extract Gibbs-ringing artifacts.
• Extracted artifacts were subtracted from original MR images.
• Low-frequency k-space data of processed images were replaced with measured data for enhanced fidelity.

Main Results:

• The CNN-based method effectively removed Gibbs-ringing artifacts in T2W and diffusion-weighted images without noticeable smoothing.
• Quantitative analysis showed improved image quality compared to state-of-the-art methods, with lower RMSE and higher PSNR and SSIM.
• The method demonstrated robust performance across various MRI sequences and anatomical regions, including brain and knee images.

Conclusions:

• The proposed CNN-based approach offers a novel and effective solution for Gibbs-ringing artifact reduction in MRI.
• The method is computationally efficient and simple to implement.
• This technique holds high applicability for routine clinical MRI, improving diagnostic accuracy.