Fig. 4From: Automatic segmentation of fat metaplasia on sacroiliac joint MRI using deep learningRepresentative examples of wrong segmentation of fat metaplasia on SIJ MRI by the 2.5D-AttentionUnet segmentation model and radiologists. a In a 28-year-old female with axSpA, normal yellow bone marrow fat of the sacrum and normal intermuscular fat were incorrectly segmented by the model and radiological resident (white arrows). b In a 17-year-old male with axSpA, the model incorrectly segmented normal fat in the sacroiliac joint space (white arrow) rather than radiologists. c In a 27-year-old female with non-axSpA (non-specific sacroiliitis), the model and radiologists incorrectly segmented the normal yellow bone marrow of the sacrum and ilium (white arrows). d In a 39-year-old female with axSpA, small patchy FMs in the sacrum and ilium (white arrows) were missed by two radiologists, which are segmented by the model. axSpAÂ axial spondyloarthritis; non-axSpAÂ non-axial spondyloarthritis; T1WIÂ T1-weighted images; DLÂ deep learning; DSCÂ Dice similarity coefficientBack to article page