Quantification of fat in the posterior sacroiliac joint region applying a semi-automated segmentation method

Background and objective Fat within the posterior sacroiliac joint region (PSIJ) is thought to compensate for the incongruent surfaces of the sacrum and ilium posteriorly. Knowledge on the presence of fat in the SIJ could provide useful information about joint physiology and clinical kinematic implications of its presence. This study aimed at quantifying fat within the PSIJ, using a semi-automated method, and to compare the results to a manual segmentation method based on data from frozen cadaveric sections and computed tomography (CT). The results may provide a quicker and more objective method for fat volume quantification. Methods Seventy-eight cadaveric hemipelves were used. Frozen sections were obtained and photographed and CT data obtained from subsamples. A MATLAB routine was deployed to assess fat in the serial sections and CT scans, using masks derived from color thresholds and Hounsfield units, respectively. Regions of interest were created to isolate the PSIJ region before fat volume was computed. A Friedman test was used for the comparison between all masks and the manual method, a Kruskall-Wallis test for comparing the CT results with all masks and the manual method and Bland-Altman plots were used to express the result differences of these methods. Results PSIJ fat volume averaged 3.9 ± 2.2, 4.9 ± 2.5, 3.7 ± 2.3 and 7.2 ± 7.3 cm3 for masks 1 (fat mask), 2 (no-fat mask), 3 ('control' fat mask) and CT, respectively. All masks and the CT fat volume were significantly different to the manual segmentation method (p<0.01). Mask 2 differed significantly from masks 1 and 3 (both p<0.01). Bland-Altman plots yielded differences in the measurements between the various methods. Conclusions Manual segmentation of PSIJ fat volume may result in a relative underestimation of the total fat compared to semi-automated or CT-based methods, as fat might not be sufficiently distinguished from surrounding structures. However, the CT-based method resulted in vastly higher variation in the results and warrants further study. The semi-automated approach to quantify fat based on color thresholds presented here is more investigator-independent, time efficient and applicable to CT scans, which provides opportunity to use this technique on various tissue types in vivo.