Effects of patient and imaging factors on small bowel motility scores derived from deep learning-based segmentation of cine MRI

초록

Objectives Small bowel motility can be quantified using cine MRI, but the influence of patient and imaging factors on motility scores remains unclear. This study evaluated whether patient and imaging factors affect motility scores derived from deep learning-based segmentation of cine MRI. Materials and methods Fifty-four patients (mean age 53.6 +/- 16.4 years; 34 women) with chronic constipation or suspected colonic pseudo-obstruction who underwent cine MRI covering the entire small bowel between 2022 and 2023 were included. A deep learning algorithm was developed to segment small bowel regions, and motility was quantified with an optical flow-based algorithm, producing a motility score for each slice. Associations of motility scores with patient factors (age, sex, body mass index, symptoms, and bowel distension) and MRI slice-related factors (anatomical location, bowel area, and anteroposterior position) were analyzed using linear mixed models. Results Deep learning-based small bowel segmentation achieved a mean volumetric Dice similarity coefficient of 75.4 +/- 18.9%, with a manual correction time of 26.5 +/- 13.5 s. Median motility scores per patient ranged from 26.4 to 64.4, with an interquartile range of 3.1-26.6. Multivariable analysis revealed that MRI slice-related factors, including anatomical location with mixed ileum and jejunum (beta = -4.9; p = 0.01, compared with ileum dominant), bowel area (first order beta = -0.2, p < 0.001; second order beta = 5.7 x 10(-4), p < 0.001), and anteroposterior position (first order beta = -51.5, p < 0.001; second order beta = 28.8, p = 0.004) were significantly associated with motility scores. Patient factors showed no association with motility scores. Conclusion Small bowel motility scores were significantly associated with MRI slice-related factors. Determining global motility without adjusting for these factors may be limited.

키워드