Belly Fat on CT Scans Shows Hidden Clues That Separate Crohn’s Disease From Appendicitis

When a patient arrives at the emergency department with pain in the lower right abdomen, doctors face one of medicine’s most familiar diagnostic puzzles: is it acute appendicitis, or could it be Crohn’s disease inflaming the same stretch of bowel? Both conditions light up the same region on a CT scan, both produce that hazy streaking of fat around the intestine known as fat stranding, and both can mimic each other closely enough that even experienced radiologists hesitate. A new exploratory study from researchers at Wannan Medical University and Tongling Municipal Hospital in Anhui, China, has now taken a quantitative approach to this problem, measuring ten objective indices from ordinary noncontrast CT scans to see whether numbers alone can separate the two diseases. The findings, published in BMC Medical Imaging, are both promising and sobering: the two conditions do differ measurably, but they overlap far more than many clinicians might expect.

The study, led by first authors Xiao Hu and Nian-Xia Qian under the correspondence of Xiao-Dong Liu, enrolled 49 patients with active, mild-to-moderate ileocecal Crohn’s disease and 78 patients whose acute appendicitis was confirmed by histopathology after surgery. Rather than relying on the subjective visual impression of inflammation, the team extracted precise measurements from each scan. These included the density of fat immediately surrounding the diseased bowel segment, the density of subcutaneous fat beneath the skin, the density of fat at the root of the mesentery, the attenuation of the psoas muscle in the lower back, the height of the peritoneal cavity, overall abdominal height, and four derived ratios labeled rCT1, rCT2, rCT3, and the peritoneal-to-abdominal height ratio, or PAR. Two radiologists measured every index independently and their values were averaged, a design that helps guard against the idiosyncrasies of any single observer.

The underlying logic is elegant. Fat density, expressed in Hounsfield units on the CT scale, rises when fat becomes inflamed and edematous, because fluid infiltration makes the tissue appear denser than healthy, airy adipose tissue. Muscle attenuation, meanwhile, reflects overall body composition and nutritional state, which chronic inflammatory diseases like Crohn’s can erode over time. Abdominal geometry, captured through peritoneal and abdominal height and their ratio, may reflect differences in body habitus and in the way disease redistributes tissue within the abdomen. By combining these simple, reproducible measurements, the researchers hoped to build a phenotypic fingerprint of each disease that no single scan reading could provide.

The results revealed genuine biochemical-visual differences between the groups. Perilesional fat density, the fat hugging the inflamed bowel, was lower in Crohn’s disease than in appendicitis, averaging minus 77.4 Hounsfield units versus minus 68.1. That counterintuitive direction suggests that the chronic, fibrofatty changes of Crohn’s disease, often described as creeping fat, alter the surrounding tissue differently from the acute, water-laden inflammation of a suddenly obstructed appendix. In contrast, the peritoneal-to-abdominal height ratio was higher in Crohn’s patients, at 0.55 versus 0.50, and psoas muscle attenuation was also higher, at 55.1 versus 50.1 Hounsfield units, hinting at systematic differences in body composition and abdominal architecture between the two populations.

To test whether these numbers could actually tell the diseases apart, the team built a three-index logistic regression model combining rCT2, PAR, and psoas muscle attenuation. In the two-group comparison, the model achieved an apparent area under the receiver operating characteristic curve of 0.880, with a 95 percent confidence interval of 0.82 to 0.94, and a leave-one-out cross-validated AUC of 0.854. Cross-validation matters here: by repeatedly holding out one patient and retraining on the rest, the LOOCV figure gives a more honest estimate of how the model behaves on unseen data than the apparent AUC alone. Still, the authors are careful to frame these values as quantifying phenotypic separation between the two groups rather than as diagnostic accuracy in a clinical sense, an important distinction that separates this study from the many overconfident machine-learning papers in the medical imaging literature.

Perhaps the most revealing analysis, however, was the one-class approach. Instead of comparing the two groups head to head, the researchers defined a 95 percent reference region using only the appendicitis patients, then asked how many Crohn’s patients fell outside that region. If the two diseases were truly distinct phenotypes, most Crohn’s cases should have landed outside. In reality, only 20.4 percent of Crohn’s patients, 10 of 49, fell outside the appendicitis-derived reference region, while just 2.6 percent of appendicitis cases, 2 of 78, did so. In other words, roughly four out of five Crohn’s patients looked, by these quantitative measures, statistically indistinguishable from appendicitis patients. That is a striking degree of overlap for two conditions with fundamentally different underlying biology, and it explains why visual differentiation on CT remains so difficult.

The team then stress-tested the transportability of their reference region using a separate cohort of 89 appendicitis cases, 59 from a different scanner at the same institution and 30 from an external center. Among the same-institution, different-scanner cases, 100 percent fell within the 95 percent reference region, a reassuring result for internal consistency. But among the external-center cases, only 80 percent, 24 of 30, fell inside. That 20 percent failure rate across centers is a familiar warning in radiology research: CT numbers are not universal constants. Differences in scanner manufacturer, calibration, reconstruction algorithms, and radiation dose can shift Hounsfield unit measurements enough to undermine models trained at a single site. No independent Crohn’s disease cohort was available for external validation, which the authors acknowledge as a key limitation.

Methodologically, the study is notable for its transparency and restraint. The researchers report following STROBE guidelines for observational studies and TRIPOD+AI guidelines for prediction models, and their supplementary material includes principal component analysis, hierarchical clustering, ROC curves, calibration plots, and sensitivity analyses. They also paid attention to the events-per-variable problem, a common pitfall in small clinical prediction studies where too many predictors are fit to too few outcomes, inflating apparent performance. By limiting their final model to three indices and reporting cross-validated rather than apparent performance, they avoided some of the most frequent statistical traps. The work was supported by the Natural Science Research Project of the Anhui Provincial Department of Education, and the ethics committee of Tongling Municipal Hospital approved the retrospective design with a waiver of written informed consent.

What does this mean for patients and clinicians today? The authors are explicit: the substantial overlap between the two conditions limits the use of these CT indices as a standalone discriminator, and the findings are exploratory and should not be interpreted as a clinically validated diagnostic tool. A patient with lower right abdominal pain should not expect a blood test or a scan measurement to reliably settle the Crohn’s-versus-appendicitis question anytime soon. Clinical context, laboratory findings such as white blood cell count, symptom duration, and follow-up imaging with CT enterography or magnetic resonance enterography remain the backbone of diagnosis. What the study does offer is a rigorous, honest baseline: it shows which quantitative features carry signal, how much they overlap, and how fragile such measurements can be when they cross institutional boundaries.

The broader significance lies in the study’s framing of a problem that will only grow as artificial intelligence enters radiology. Hundreds of papers each year report models that distinguish diseases on scans with impressive AUCs, yet few ask the harder question of whether the measured phenotype is stable across scanners, centers, and populations. By publishing a one-class reference analysis, an external-scanner cohort, and a candid account of failure modes, the Anhui team has modeled the kind of self-skeptical science that medical imaging needs. The creeping fat of Crohn’s disease and the fiery inflammation of appendicitis may never be fully separable by numbers alone, but knowing exactly where the boundary blurs is the first step toward tools that genuinely help the radiologist staring at that ambiguous lower right quadrant.

Subject of Research: Quantitative noncontrast CT indices of perienteric fat and abdominal geometry in ileocecal Crohn disease versus acute appendicitis

Article Title: Noncontrast CT indices of perienteric fat and abdominal geometry in ileocecal Crohn disease and acute appendicitis: an exploratory comparison

Article References: Hu, X., Qian, N.-X., Chang, L.-H., Ren, Y.-Q., Gao, T., & Liu, X.-D. (2026). Noncontrast CT indices of perienteric fat and abdominal geometry in ileocecal Crohn disease and acute appendicitis: an exploratory comparison. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02901-3

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02901-3

Keywords: Crohn's disease, acute appendicitis, computed tomography, fat stranding, perienteric fat density, Hounsfield units, psoas muscle attenuation, peritoneal-to-abdominal height ratio, radiology, diagnostic imaging, inflammatory bowel disease, cross-validation