Differentiating multilocular hepatic cysts from mucinous cystic neoplasms: characteristic imaging signs and a machine learning diagnostic framework.

Background: Differentiating benign multilocular hepatic cysts (MHCs) from premalignant mucinous cystic neoplasms (MCNs) is crucial for management but remains challenging due to overlapping imaging features. This study aimed to enhance diagnostic accuracy in distinguishing MHCs from MCNs by evaluating morphological imaging characteristics and constructing a predictive diagnostic model. Methods: This retrospective study included a cohort of 86 patients with pathologically confirmed hepatic cystic lesions, 60 with MHCs and 26 with MCNs, diagnosed between July 2014 and April 2023. More than 20 imaging features were independently assessed by two radiologists blinded to clinical history. Feature selection was performed using Boruta regression, least absolute shrinkage and selection operator (LASSO) regression, as well as univariate and multivariate logistic regression analyses. Diagnostic performance was assessed across four machine learning models: logistic regression, random forest (RF), decision tree, and extreme gradient boosting (XGBoost). Firth logistic regression was used to examine the relationship between the ribbon sign and intracystic hemorrhage. Results: Multiple lesions, smooth septal appearance, peripheral septal location, and the ribbon sign were more suggestive of MHCs, whereas solitary lesions, irregular septa, cyst wall thickening (≥2 mm), and the septum-intersection triangular sign were more indicative of MCNs. Among the models, the XGBoost classifier presented the highest diagnostic performance [area under the curve (AUC) =0.905]. The ribbon sign was significantly associated with intracystic hemorrhage (odds ratio =608.46, P<0.001), with no significant interaction observed between this association and lesion type (P=0.272). Conclusions: A high-performance diagnostic model was developed using XGBoost by integrating advanced feature selection techniques and machine learning algorithms. The ribbon sign was identified as a significant imaging marker independently associated with intracystic hemorrhage, regardless of lesion type.