CT prediction of resectability and prognosis in patients with pancreatic cancer after neoadjuvant treatment using image findings and texture analysis

Abstract

Purpose: To assess the utility of CT findings and texture analysis for predicting resectability and prognosis after neoadjuvant therapy in patients with surgery for pancreatic cancer. Materials and Methods: From 2013 to 2016, among 308 patients, forty-five patients with pancreatic cancer underwent both neoadjuvant therapy and surgery were included. They underwent neoadjuvant concurrent chemoradiation therapy (CCRT, n=27) or neoadjuvant chemotherapy (ChoT, n=18). All patients performed baseline and preoperative CT. Two reviewers assessed CT findings and resectability (resectable, borderline resectable, unresectable). Residual tumor categorized into no residual tumor (R0) and residual tumor (R1 or R2). We analyzed the relationship between CT findings and R classification. CT texture analysis was performed by PC-based in-house software using baseline and preoperative CT. Texture values obtained by subtracting preoperative CT from baseline CT were analyzed using multivariate Cox/logistic regression analysis to identify significant parameter for prediction of resectability and prognosis. Results: There were 30 patients without residual tumor (CCRT, n=20

ChoT, n=10) and 15 patients with residual tumor (CCRT, n=7

ChoT, n=8). Overall accuracy for R0 resectability was better when considering borderline as resectable tumor [68.9% (31 of 45) for both readers] than when considering borderline as unresectable tumor [55.6% (25 of 45) for reader 1 and 51.1% (23 of 45) for reader 2] (P<0.001). Considering borderline as resectable tumor, CCRT group has better accuracy than chemotherapy group (77.8% vs 55.6%, P=0.545 for reader1, 74.1% vs 61.1% P=0.279 for reader 2). On the contrary, considering borderline as unresectable tumor, chemotherapy group has better accuracy than CCRT group (61.1% vs 51.9% P=0.119 for reader1, 61.1% vs 44.4% P=0.363 for reader 2). In CT texture analysis, three subtracted texture values were found to be independent predictors of R0 resection

surface area (OR 1.077, P=0.011), GLCM IDM (OR 0.000, P=0.005) and GLCM contrast (OR 0.982, P=0.012) and two subtracted texture values were associated with overall survival

entropy (HR 0.159, P=0.005) and GLCM entropy (HR 10.235, P=0.036). Conclusion: After neoadjuvant therapy, considering borderline as resectable tumor have better accuracy for R0 resectability, particularly in CCRT group and CT with texture analysis can be useful to predict patients outcome after neoadjuvant therapy in pancreatic cancer.