Deep learning enabled motion detection in quantitative macromolecule proton faction mapping in the liver

Motivation: Macromolecular proton fraction (MPF) quantification based on spin-lock (MPF-SL) is a technology, which sensitively measures macromolecule content using the MT effect. However, the motion of liver can lead to inaccurate MPF-SL quantification. Goal(s): Develop an automated processing approach that can detect motion of the liver to reduce the impact on MPF-SL quantification. Approach: We trained a deep learning model to automatically detect the motion of the liver during MPF-SL acquisition. Results: The proposed model demonstrated good performance with an accuracy of 86.4% and an area under the receiver operating characteristic curve (AUC) of 0.79. Impact: Our approach enables automated motion detection of the liver during MPF-SL scan. It can improve reliability of parameter quantification by either discarding unreliable measurements retrospectively or prompt data recollection prospectively during scanning.