OBJECTIVE. The objective of our study was to identify relevant and reliable
CT signs of bowel injury, to determine the overall performance of CT in de
tection of bowel injuries, and to establish the effect of the training leve
l of radiologists on this performance.
MATERIALS AND METHODS. Abdominal CT scans of 112 patients with blunt abdomi
nal trauma were prospectively and retrospectively reviewed. Fifty patients
had proven bowel injuries (with or without other visceral injuries), wherea
s 62 patients had no bowel injury and comprised the comparison or control g
roup. Thirty-one of the 62 patients in the comparison group had surgical pr
oof of abdominal but not bowel or mesenteric injuries. The retrospective re
view of the 112 CT scans was performed randomly and individually by nine ra
diologists unaware of the diagnosis, including three faculty abdominal radi
ologists, three senior residents in training, and three junior residents in
training. Individual performance and group performance were evaluated by r
eceiver operating characteristic analysis, and interobserver agreement was
tested. Individual CT signs as relevant predictors of bowel injury were ide
ntified by logistic regression.
RESULTS. Relevant predictors of bowel injury included mesenteric infiltrati
on, bowel wall thickening, extravasation of vascular or enteric contrast ag
ent, and the presence free air. In the retrospective blinded review, CT sho
wed good to excellent interobserver reliability for individual CT signs as
well as for diagnosis of bowel and visceral injuries. Faculty radiologists
tended to diagnose injuries with greater accuracy and confidence, but they
showed significantly better performance than residents only in diagnosing d
uodenal perforation. For the prospective CT diagnosis of bowel injury, CT h
ad a sensitivity of 64%, an accuracy of 82%, and a specificity of 97%.
CONCLUSION. Rowel injuries are challenging to diagnose on CT. Radiologists
with various levels of experience and expertise can achieve accurate and re
producible results using a variety of CT criteria.