Radiomics-based prediction of recurrent acute pancreatitis in individuals with metabolic syndrome using T2WI magnetic resonance imaging data.

OBJECTIVE

This study sought to clarify the utility of T2-weighted imaging (T2WI)-based radiomics to predict the recurrence of acute pancreatitis (AP) in subjects with metabolic syndrome (MetS).

METHODS

Data from 196 patients with both AP and MetS from our hospital were retrospectively analyzed. These patients were separated into two groups according to their clinical follow-up outcomes, including those with first-onset AP (n = 114) and those with recurrent AP (RAP) (n = 82). The 196 cases were randomly divided into a training set (n = 137) and a test set (n = 59) at a 7:3 ratio. The clinical characteristics of these patients were systematically compiled for further analysis. For each case, the pancreatic parenchyma was manually delineated slice by slice using 3D Slicer software, and the appropriate radiomics characteristics were retrieved. The K-best approach, the least absolute shrinkage and selection operator (LASSO) algorithm, and variance thresholding were all used in the feature selection process. The establishment of clinical, radiomics, and combined models for forecasting AP recurrence in patients with MetS was then done using a random forest classifier. Model performance was measured using the area under the receiver operating characteristic curve (AUC), and model comparison was done using the DeLong test. The clinical utility of these models was evaluated using decision curve analysis (DCA), and the optimal model was determined via a calibration curve.

RESULTS

In the training set, the clinical, radiomics, and combined models yielded respective AUCs of 0.651, 0.825, and 0.883, with corresponding test sets of AUCs of 0.606, 0.776, and 0.878. Both the radiomics and combined models exhibited superior predictive effectiveness compared to the clinical model in both the training (p = 0.001, p < 0.001) and test sets (p = 0.04, p < 0.001). The combined model outperformed the radiomics model (training set: p = 0.025, test set: p = 0.019). The DCA demonstrated that the radiomics and combined models had greater clinical efficacy than the clinical model. The calibration curve for the combined model demonstrated good agreement between the predicted probability of AP recurrence and the observed outcomes.

CONCLUSION

These findings highlight the superior predictive power of a T2WI-based radiomics model for predicting AP recurrence in patients with MetS, potentially supporting early interventions that can mitigate or alleviate RAP.