Detailed Abstract
[Plenary Session - Biliary & Pancreas (Pancreas Disease/Surgery)]
[PL-S3] Prediction of Lymph Node Metastasis in Pancreatic Cancer with Arterial And Venous Phase of Contrast Enhanced Computed Tomography Images Using Radiomics Analysis
Ji Su KIM 1, Hyein CHO 2, Hwiyoung KIM 2, Chang Moo KANG 3
1 Division of Hepatobiliary, Pancreas, And Abdominal Organ Transplant, Department of Surgery, Incheon St. Mary's hospital, The Catholic University of Korea, REPUBLIC OF KOREA, 2 Department of Biomedical Systems Informatics, Severance Hospital, REPUBLIC OF KOREA, 3 Division of HBP Surgery, Severance Hospital, REPUBLIC OF KOREA
Background : The objective of this study was to construct a radiomics-based model using arterial (A) and venous (V) phases of contrast enhanced computed tomography (CECT) images to predict pathological lymph node (LN) metastasis in resected pancreatic cancer(PC).
Methods : We retrospectively collected preoperative CECT images from 208 PC patients who underwent surgical resection between October 2011 and December 2020. The dataset consisted of 137 A images and 71 V images. A total of 1,022 radiomics features were extracted from each phase, and the most important features were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) method. The selected significant features were then classified using Support Vector Machine (SVM) and Ridge regression models. Long-term oncologic outcomes were compared between actual pathological LN metastasis groups and radiomics-based predicting groups.
Results : The LASSO identified significant radiomics features associated with LN metastasis in PC. The AUC values for the Ridge classifiers were 0.94 (A) and 0.91 (V), while the AUC values for the SVM classifiers were 0.90 (A) and 0.88 (V). Disease-free survival (median 21.2 vs. 10.4 months, p<0.0001), and disease-specific survival (median 46.07 vs. 25.57 months, p=0.008) were statistically different according to radiomics-based LN prediction groups, which were no statistically different from those according to pN groups (p>0.05).
Conclusions : This study demonstrates that the radiomics based model using arterial and venous phases of CECT images not only predict actual LN metastasis, but also oncological correlated with long-term survival outcomes in resected PC. This can provide potential rationales in determining neoadjuvant chemotherapy in resectable PC.
Methods : We retrospectively collected preoperative CECT images from 208 PC patients who underwent surgical resection between October 2011 and December 2020. The dataset consisted of 137 A images and 71 V images. A total of 1,022 radiomics features were extracted from each phase, and the most important features were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) method. The selected significant features were then classified using Support Vector Machine (SVM) and Ridge regression models. Long-term oncologic outcomes were compared between actual pathological LN metastasis groups and radiomics-based predicting groups.
Results : The LASSO identified significant radiomics features associated with LN metastasis in PC. The AUC values for the Ridge classifiers were 0.94 (A) and 0.91 (V), while the AUC values for the SVM classifiers were 0.90 (A) and 0.88 (V). Disease-free survival (median 21.2 vs. 10.4 months, p<0.0001), and disease-specific survival (median 46.07 vs. 25.57 months, p=0.008) were statistically different according to radiomics-based LN prediction groups, which were no statistically different from those according to pN groups (p>0.05).
Conclusions : This study demonstrates that the radiomics based model using arterial and venous phases of CECT images not only predict actual LN metastasis, but also oncological correlated with long-term survival outcomes in resected PC. This can provide potential rationales in determining neoadjuvant chemotherapy in resectable PC.
SESSION
Plenary Session
Room A 3/23/2024 11:20 AM - 12:20 PM