Developing a CT-based radiomics nomogram for predicting post-acute pancreatitis diabetes mellitus incidence

The present study aimed to develop the utility of a nomogram based on clinical and radiomics as a tool for predicting post-acute pancreatitis diabetes mellitus (PPDM-A).

This retrospective investigation evaluated 244 patients with acute pancreatitis. Patients were randomized in a 7:3 ratio into training and validation cohorts. Radiomics feature selection was then achieved using the variance threshold, select best K, and least absolute shrinkage and selection operator methods. The area under the curve values, decision, and calibration curves have been used to determine the models' predictive value.

The developed nomogram performed superior to the clinical model in the validation (0.815 vs 0.677, p = 0.016) and training cohorts (0.803 vs 0.683, p = 0.002). The calibration curves demonstrated that the expected and actual values were satisfactory. In contrast, decision curve analysis revealed a stronger relationship between the nomogram and net clinical value than with the distinct radiomics or clinical signature effects.

In summary, the findings of this study demonstrated that establishing a predictive nomogram as a non-invasive technique may be useful in predicting the risk of PPDM-A.

This is the first time to use a CT radiomics nomogram to predict PPDM-A. The nomogram is conducive to the personalized prediction of patients. It only needs to input the patient's information, and a simple addition operation can quantitatively obtain its risk. The resultant tool has the potential to provide new opportunities to treat or prevent PPDM-A more effectively.