Development of Automated Segmentation of the Thigh Muscles from Dixon MRI for Fat Fraction Quantification

Abstract

Sarcopenia, an age related condition is associated with decreased in lean muscle mass and increased in muscle fat. As a result, there is a decline in muscle strength and function in older adults. Diagnosis can be performed using bio-electrical impedance analysis (BIA) though accuracy is influenced by factors such as age, gender, hydration and ethnicity. Image based biomarkers have emerged as potentially more objective for diagnosis and from several possible modalities, MRI is considered a reference. In particular multiecho sequences for chemical shift imaging such as the Dixon method can be used to enhance muscle and fat contrast and calculate intramuscular fat (IMF) infiltration. In image based analysis, there is a need for automation, specifically in the segmentation of the muscles. In this work, we propose an automatic segmentation pipeline for a multipoint Dixon sequence. We evaluated the method with a publicly available dataset and compared it with the ground truth. We also demonstrated the method using local data from an MRI scan of the thigh muscles of an older person. The results showed that the mean PDFF of the right thigh muscle segmented by the proposed method correlates well with the mean PDFF from the ground truth segmentation, with a correlation value of 0.877. Qualitatively, the proposed method also produced a good segmentation of our local data. This suggests that the proposed method of muscle fat quantification can be used in future studies.