435 Background: Pancreatic cancer patients with liver metastases had much poorer prognosis than those with other metastatic patterns. This study aimed to develop and validate a radiomics model to discriminate pancreatic cancer patients with liver metastases from patients with other metastatic patterns. Methods: We evaluated 77 patients advanced pancreatic cancer (APC) with different metastatic patterns and performed texture analysis on the region of interest (ROI). 58 patients and 19 patients were allocated randomly into the training cohort and the validation cohort with almost the same proportion of patients with liver metastases. An independent samples t-test was used for initial feature selection in the training cohort. Random Forest Classifier (RFC) was used to construct models based on these features in both cohorts and a radiomics signature (RS) was derived from the model. Then a nomogram was constructed based on RS and CA19-9, and validated with calibration plot and decision curve. The prognostic value of RS was evaluated by Kaplan-Meier methods. Results: A nomogram based on the RS and CA19-9 was constructed and it demonstrated good discrimination in the training cohort (AUC = 0.93) and validation cohort (AUC = 0.81). Kaplan-meier methods showed that patients with RS>0.61 had much poorer OS than patients with RS < 0.61 in both cohorts. Conclusions:This study presents a radiomics nomogram incorporating both RS and CA19-9, which can be used to discriminate advanced pancreatic cancer patients with liver metastases from patients with other metastatic patterns.
Time to think: Selecting patients who may benefit from synchronous resection of primary pancreatic cancer and liver metastases