Xanthogranulomatous cholecystitis (XGC) is a rare and chronic inflammatory lesion of the gallbladder characterised by special types of granulomas and foamy cells (1). Epidemiological data from previous studies showed that the incidence of XGC was 1.3%–1.9% and that in India was 8.8%, with little difference in sex among regions (2). Goodman and Ishak described XGC as a special pathological entity in 1981 (3). Although XGC is benign, it is locally aggressive and may affect nearby organs such as the liver, duodenum, colon, and common bile duct (4). It may be misdiagnosed as gallbladder cancer (GBC) or may coexist with GBC because the inflammatory process involved resembles that of GBC in a macroscopic view.

The clinical manifestations and biological examination of XGC are nonspecific and similar to acute or chronic cholecystitis (5). Several studies have reported some imaging features that highly suggest XGC, including diffuse thickening of the gallbladder wall, continuous mucosal lines, intramural hypoattenuating nodules, and the presence of gallstones in the background of chronic gallbladder disease (6, 7). Approaches based on image analysis to guide the most appropriate treatment decisions have been intensively studied (8), and the applications of artificial intelligence (9), machine learning (10), and deep learning (11) in radiology images have been increasingly explored. Imaging tools such as computed tomography (CT) and magnetic resonance imaging (MRI) are generally utilised for clinical evaluation of gallbladder diseases, so analysis of their unique radiological features through machine learning may be a better method to distinguish XGC from GBC (12).

Laparoscopic cholecystectomy is the current gold standard for the treatment of benign gallbladder diseases. However, the high conversion rate of laparoscopic cholecystectomy to open cholecystectomy in XGC is related to the difficulty in distinguishing GBC (13). Moreover, surgeons may perform improper surgery owing to the misdiagnosis of GBC (14). In this study, we retrospectively analysed 60 cases of XGC and 93 cases of GBC confirmed by histopathology to establish a diagnostic prediction model for XGC and GBC based on CT and MRI features to help clinical decision-making on the appropriate treatment method for patients with XGC.

XGC was originally considered a malignant disease process; however, many studies have now confirmed its benign course (15), which is characterised by a focal or diffuse destructive inflammatory process, followed by significant macrophage and foam cell infiltrations and hyperplastic fibrosis. Concerning the pathogenesis of XGC, the most widely accepted theory is that the extravasation of bile through the ruptured Rokitansky-Aschoff sinus or increased internal pressure of the biliary tract causes small mucosal ulcers (16). Extensive inflammatory fibrosis results in the thickening of the gallbladder wall and the formation of multiple yellowish-brown nodules (17). The inflammatory process often spreads to adjacent organs, thereby forming a tumour-like mass around the gallbladder (18). Since XGC is often misdiagnosed as GBC, which leads to unnecessary radical surgical excision, it is important to distinguish between XGC and GBC before surgery.

Although ultrasound is a clinically preferred and ubiquitous imaging method, its specificity for XGC is low (12). Clinically, CT and MRI are widely used to distinguish between XGC and GBC because of their convenience and non-invasiveness (19). We reviewed a large amount of literature in the early stages and incorporated the reported potential radiological features that may distinguish XGC from GBC into the initial model. Using the RF algorithm and Logistic regression method, it was shown that the diagnostic prediction model had a stronger ability to distinguish between XGC and GBC with an AUC of 0.888 and an accuracy of 0.898. This model, constructed with outstanding clinical applicability, summarised 11 radiological features and had high diagnostic performance, thereby confirming the importance of imaging features in the differential diagnosis of XGC and GBC.

CT findings included gallstones, mucosal lines, peripheral lymph nodes, bile duct dilation, and intramural nodules. Studies have shown that the development of XGC may be related to gallstones (17, 20), similar to our findings ( Figure 3A ). Additionally, intramural hypodense nodules, continuous mucosal lines, and absence of intrahepatic bile duct dilatation suggested XGC ( Figures 3B, C ). Goshima et al. reported that three of the five CT features that meet the XGC characteristics can provide high accuracy in distinguishing XGC from GBC (21), which is consistent with our results. Peripheral lymphadenopathy (>10 mm) was an independent factor in the diagnosis of GBC ( Figure 3D ), which may help in the correct preoperative diagnosis of XGC. As for the MRI features, the high-signal area of the T2WI corresponded to the area of significant xanthogranulomas ( Figure 3E ), which is consistent with a previous finding (22). The lipid signal of XGC nodules was owing to tissue cell phagocytosis and accumulation of extravasated bile and bile lipids. Ogawa et al. showed that the positive signal rate with DWI was significantly higher in GBC than in benign gallbladder diseases (23). DWI and ADC may play a significant role in distinguishing XGC from GBC because they are beneficial for distinguishing malignant and benign lesions.

XGC should be operated immediately once diagnosed to prevent complications owing to its invasiveness and destructiveness (24). Compared with open surgery, laparoscopic cholecystectomy has a shorter operating time and shorter postoperative hospital stay (25), which indicates that in some strictly selected cases, laparoscopic surgery is superior to open surgery for XGC treatment (4). Our predictive diagnosis model can distinguish XGC from GBC through adequate preoperative evaluation, helps clinical decision-making, and reduces the incidence of postoperative complications. Despite the high accuracy of the predictive diagnosis model, intraoperative diagnosis is still necessary because XGC may coexist with GBC. The frequency of coexistence of these two lesions was approximately 19.6% in India (26). However, there was no coexistence of XGC and GBC in our study, which is one of the limitations of this study. Owing to the retrospective study design, patients with only a single examination and obvious imaging findings were not included, so there may be an inherent selection bias. On the other hand, there were individual variables that have reported differences; however, they were not significant in our study cohort, indicating that more patients may need to be introduced in future studies.

We established a diagnostic prediction model for XGC using the RF method and logistic regression. The model contains 11 important critical variables, including T2WI signal of intramural nodules, T1WI signal of intramural nodules, lipid signal, gallstones, mucosal lines, ADC, peripheral lymph nodes, DWI, T2WI signal of the thickened cyst wall, bile duct dilation, and intramural nodules. This diagnostic prediction model has exhibited good discrimination between XGC and GBC, thus providing significant assistance in clinical practice.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by The Second Affiliated Hospital of Zhejiang University School of Medicine (Approval numbers 20210806). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

QZ conceived the study and drafted the manuscript. Ri-Sheng Yu and JX helped critically revise the manuscript for important intellectual content. QZ, CL and JZ performed the research. JY, YW and XW helped analyse data. All authors contributed to the article and approved the submitted version.

This study was supported by grants from the Zhejiang Medical and Health Science and Technology Plan Project (Grant Numbers 2019KY476).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.