Image quality assessment for MRI: Is it up to the task?

The accuracy of MRI-based diagnosis can be degraded by artifacts, a challenging problem for both radiologists and automated computer-aided systems. Assessment of MR images quality is thus of paramount importance for clinical and research purposes. In this brief review, I discuss image quality assessment (IQA) methods for MRI. After briefly describing some typical artifacts encountered in MRI, a succinct summary of popular IQA metrics is provided. A particular emphasis is put on the relevance of current IQA metrics, borrowed from existing IQA methods for natural images, when dealing with MR images with diverse contrasts and artifact types. In the IQA process, what matters for clinicians is not whether an MR image is beautiful or not but whether the clinically-relevant diagnostic pattern in the MR image is unaffected by artifacts. I then discuss the growing interest in AI-based as well as brain-inspired IQA methods, including their strengths and limitations. The possibility of integrating IQA metrics within AI image reconstruction and processing tools will have major ramifications on IQA for MRI. Challenges and promises are finally discussed in the light of the recent trends in scanning patients at ultrahigh (≥7 Tesla) or at ultralow (≤0.1 Tesla) magnetic fields with portable MRI devices.