Spectral CT radiomics features of the tumor and perigastric adipose tissue can predict lymph node metastasis in gastric cancer.

OBJECTIVES: To develop a nomogram based on the radiomics features of tumour and perigastric adipose tissue adjacent to the tumor in dual-layer spectral detector computed tomography (DLCT) for lymph node metastasis (LNM) prediction in gastric cancer (GC). METHODS: A retrospective analysis was conducted on 175 patients with gastric adenocarcinoma. They were divided into training cohort (n = 125) and validation cohort (n = 50). The radiomics features from the tumour and perigastric fat based on DLCT spectral images were extracted to construct radiomics models for LNM prediction using Lasso-GLM method. Preoperative clinicopathological features, DLCT routine parameters, and the optimal radiomics models were analyzed to establish the clinical-DLCT model, clinical-DLCT-radiomics model and a nomogram. All models were internally validated using the Bootstrap method and evaluated using receiver operating characteristic (ROC) curve. RESULTS: The area under the ROC curve (AUC) values of optimal radiomics models based on tumour (Model 1) and perigastric fat (Model 2) were 0.923 and 0.822 in training cohort, 0.821 and 0.767 in validation cohort. The clinical-DLCT model based on Nct and ECV(ID) demonstrated an AUC value of 0.728 in training cohort and 0.657 in validation cohort. The clinical-DLCT-radiomics model and the nomogram were established by incorporating Nct, ECV(ID) and the linear predictive values of Models 1 and 2, exhibiting superior predictive efficacy with an AUC value of 0.935 in training cohort and 0.876 invalidation cohort. CONCLUSIONS: The nomogram based on Nct, ECV(ID), and the radiomics features of tumour and perigastric fat in DLCT demonstrates potential for predicting LNM in GC. This approach may contribute to the development of treatment strategies and improve the clinical outcomes for GC patients.