Objective To investigate the predictive efficacy of Delta radiomics for the patholo-gical complete remission (pCR) of pancreatic cancer after total neoadjuvant therapy (TNT).

Methods The retrospective cohort study was conducted. The clinicopathological data of 263 patients with pancreatic cancer who were admitted to Henan Provincial People′s Hospital (Zhengzhou University People's Hspital) from January 2019 to September 2024 were collected. There were 166 males and 97 females, aged (56±12)years. All patients underwent TNT. The 263 patients were randomly divided into a training set of 184 cases and a test set of 79 cases using a 7∶3 random seed count. The training set was used to construct the prediction model, and the test set was used to validate the performance of the prediction model. Observation indicators: (1) postoperative and follow‑up condi-tions; (2) imaging feature selection and model construction; (3) evaluation of predictive efficacy of different radiomic models. Comparison of measurement data with normal distribution between groups was conducted using the t test. Comparison of measurement data with skewed distribution between groups was conducted using theMann‑Whitney U test. Comparison of count data between groups was conducted using the chi‑square test. The Kaplan‑Meier method was used to calculate the survival rate and draw survival curve. The Log‑rank test was used for survival analysis. The perfor-mance of the prediction model for pCR after TNT was evaluated using the receiver operator charac-teristic (ROC) curve, precision‑recall (P‑R) curve and Bootstrap method, along with the calculation of area under the curve (AUC), precision rate, recall rate, F1-score.

Results (1) Postoperative and follow‑up conditions. All 263 patients underwent surgery after TNT, with pathological examination revealing 124 cases of pCR (86 cases in the training set, 38 cases in the test set) and 139 cases of non‑pCR (98 cases in the training set, 41 cases in the test set), respectively. All 263 patients were followed up for 6(range, 3-12) months after surgery, of which 15 cases (4 cases of pCR and 11 cases of non-pCR) were lost to follow-up or died due to non-tumor reasons within 6 months after surgery. The postoperative 6-month recurrence-free survival rate of 124 pCR patients and 139 non-pCR patients were 80% and 50%, respectively, showing a significant difference between the two groups of patients (χ²=22.84, P<0.05). (2) Imaging feature selection and model construction. Construction of the traditional radiology model: based on the response evaluation criteria in solid tumors 1.1, the Logistic regression model was constructed using the relative shrinkage (D%) as a predictive variable. The AUC of traditional radiology model was 0.72 95% confidence interval (CI) as 0.63‒0.81 in the training set and 0.75 (95%CI as 0.66‒0.84) in the test set, respectively. Construction of the Delta radiomics model: 10 non‑zero coefficient features were selected. The Delta radiomics models were constructed by using the regularized Logistic regression, random forest, gradient boosting machine, and support vector machine algorithms through using selected features as input variables. (3) Evaluation of predictive efficacy of different radiomic models. The AUC of Delta radiomics model constructed by regularized Logistic regression algorithm in the test set for predicting pCR in pancreatic cancer after TNT was 0.90, higher than that of the random forest algorithm, gradient boosting machine algorithm, support vector machine algorithm (AUC as 0.81, 0.81, 0.83), and higher than that of the traditional radiology model (AUC as 0.72). Results of Bootstrap method revealed significant differences in the predictive efficacy of Delta radiomics model constructed by regularized Logistic regression algorithm compared to the Delta radiomics model constructed by random forest algorithm, gradient boosting machine algorithm, support vector machine algorithm and the tradi-tional radiology model (95%CI as 0.03‒0.16, 0.03‒0.16, 0.03‒0.13, 0.08‒0.29, P<0.05). The regularized Logistic regression algorithm within the Delta radiomics model demonstrated the best overall performance among the above models evaluated.

Conclusion Compared to the traditional radiology model, the Delta radiomics model offers superior efficacy in predicting pCR of pancreatic cancer after TNT, in which the regularized Logistic regression algorithm demonstrates the best overall performance metrics.