Preoperative CT-based radiomics of suprapancreatic adipose tissue for predicting high-difficulty lymph node dissection in gastric cancer.
Authors
Affiliations (6)
Affiliations (6)
- Department of General Surgery, Beijing Friendship Hopital, Capital Medical University, 100050, Beijing, China.
- National Clinical Research Center for Digestive Diseases, 100050, Beijing, China.
- State Key Lab of Digestive Health, 100050, Beijing, China.
- Department of General Surgery, Beijing Friendship Hopital, Capital Medical University, 100050, Beijing, China. [email protected].
- National Clinical Research Center for Digestive Diseases, 100050, Beijing, China. [email protected].
- State Key Lab of Digestive Health, 100050, Beijing, China. [email protected].
Abstract
Preoperative assessment of lymph node dissection (LND) difficulty in gastric cancer remains challenging. Conventional clinical indicators are relatively coarse and may not adequately reflect tissue-related complexity within the surgical field. This study aimed to develop and validate a preoperative CT-based radiomics approach using suprapancreatic adipose tissue for predicting high-difficulty LND in gastric cancer. This single-center retrospective study included 192 patients with gastric cancer who underwent laparoscopic radical D2 gastrectomy between January 2022 and December 2024. Patients were randomly assigned to a training cohort (n = 134) and a validation cohort (n = 58). High-difficulty LND was defined as suprapancreatic LND time exceeding the 75th percentile (32.68 min) and/or unqualified suprapancreatic dissection according to KLASS-02-QC. A single-slice ROI was manually delineated in the suprapancreatic adipose tissue on portal venous-phase CT images. Radiomics features were extracted, followed by feature selection and model development. Four radiomics models, a clinical model, and a combined model were constructed. Sensitivity analyses were further performed using alternative time thresholds and in the subgroup without neoadjuvant chemotherapy (NCT). A total of 81 of 192 patients (42.2%) were classified as having high-difficulty LND. Nine radiomics features were finally selected to construct the radiomics signature. Among the four radiomics models, the RF model showed strong predictive performance, with an AUC of 0.877 in the training cohort and 0.855 in the validation cohort. The clinical model achieved AUCs of 0.861 and 0.804 in the training and validation cohorts, respectively. The combined model integrating radiomics and clinical features achieved the best performance, with AUCs of 0.917 in the training cohort and 0.950 in the validation cohort. Calibration and decision curve analyses also indicated favorable model performance. Sensitivity analyses showed broadly stable model performance across adjacent time-threshold definitions, whereas the selected RF radiomics model showed attenuated discrimination in patients without NCT. Preoperative CT-based radiomics of suprapancreatic adipose tissue showed potential for predicting high-difficulty LND in gastric cancer. A combined radiomics-clinical model achieved better predictive performance than either radiomics or clinical features alone and may provide a useful imaging-based framework for preoperative assessment. However, because part of the observed radiomic signal may reflect treatment-related tissue changes, particularly in patients receiving NCT, further external validation and treatment-stratified investigation are warranted.