Ringing Artifact Removal Using Zero-Shot Deep Anti-Aliasing Prior in MR image

Motivation: MRI image quality is degraded by artifacts like Gibbs ringing, BLADE aliasing, and motion distortions. Current removal methods require labeled data or cause blurring. A general solution without extensive training is needed. Goal(s): Develop an adaptable zero-shot method to remove MRI ringing artifacts across contrasts, resolutions, and scenarios without labeled data, pre-training, or explicit priors. Approach: Using the implicit prior of randomly initialized CNNs, we proposed a zero-shot artifact removal method using a CNN with an anti-aliasing prior. Results: The method suppresses Gibbs ringing, BLADE, and motion artifacts, preserving edge details. It has limitations for severe motion artifacts but performs well across contrasts. Impact: This work proposed a deep anti-aliasing prior to remove the ringing artifacts from various sources in a zero-shot manner.