Synopsis

Keywords: Image acquisition: Machine learning

This lecture will review some of the techniques to minimize the risk of unwanted artifacts in AI-based image reconstruction that can be embedded in the model development process. In addition, opportunities for automated testing of model robustness and image quality validation will be discussed.

AI-based reconstruction has the potential to significantly improve image quality compared to prior fast imaging techniques such as parallel imaging and compressed sensing. This can be used to further speed up MRI scans leading to higher patient throughput and improved motion robustness. Research in this field has shown many success stories and all major vendors offer an AI-based reconstruction product option.
However, there are concerns regarding the quality and reliability of AI-based image reconstruction. How would the reconstruction fail? Would it remove or add structure to the images? Are potential artifacts clearly recognizable or do they mimic anatomy/pathology? Questions like these are often discussed and the concerns are understandable. AI-based image reconstruction techniques are not a magic wand and like other techniques, they can also fail.
This lecture will review some of the techniques to minimize the risk of unwanted artifacts in AI-based image reconstruction that can be embedded in the model development process. In addition, opportunities for automated testing of model robustness and image quality validation will be discussed.

Acknowledgements

No acknowledgement found.