A novel semi-automatic segmentation method for volumetric assessment of the colon based on magnetic resonance imaging

To develop a novel semi-automatic segmentation method for quantification of the colon from magnetic resonance imaging (MRI).

Fourteen abdominal T2-weighted and dual-echo Dixon-type water-only MRI scans were obtained from four healthy subjects. Regions of interest containing the colon were outlined manually on the T2-weighted images. Segmentation of the colon and feces was obtained using k-means clustering and image registration. Regional colonic and fecal volumes were obtained. Inter-observer agreement between two observers was assessed using the Dice similarity coefficient as measure of overlap.

Colonic segmentations showed wide variation in volume and morphology between subjects. Colon volumes of the four healthy subjects for both observers were (median [interquartile range]) ascending colon 200 mL [169.5-260], transverse 200.5 mL [113.5-242.5], descending 148 mL [121.5-178.5], sigmoid-rectum 277 mL [192-345], and total 819 mL [687-898.5]. Overlap agreement for the total colon segmentation between the two observers was high with a Dice similarity coefficient of 0.91 [0.84-0.94]. The colon volume to feces volume ratio was on average 0.7.

Regional colon volumes were comparable to previous findings using fully manual segmentation. The method showed good agreement between observers and may be used in future studies of gastrointestinal disorders to assess colon and fecal volume and colon morphology. Novel insight into morphology and quantitative assessment of the colon using this method may provide new biomarkers for constipation and abdominal pain compared to radiography which suffers from poor reliability.