Use of radiomics to extract splenic features to predict prognosis of patients with gastric cancer.

INTRODUCTION

Radiomics allows for mining of imaging data to examine tissue characteristics non-invasively, which can be used to predict the prognosis of a patient. This study explored the use of imaging techniques to evaluate splenic tissue characteristics to predict the prognosis of patients with gastric cancer.

MATERIALS AND

METHODS

Computed tomography images from patients with gastric cancer were collected retrospectively. Splenic image characteristics, extracted with pyradiomics, of patients in the training group were randomly divided. Characteristics with a P value < 0.1 were selected for lasso regression to construct a survival risk model. Models for high-and low-risk groups were established. Patients were divided into the high- and low-risk groups for univariate and multivariate regression analysis of survival-related factors, and a visual prognostic prediction model was established.

RESULTS

The splenic characteristic prognostic model was consistent in the training and verification groups (p < 0.001 and p = 0.016, respectively). The two groups that displayed different splenic characteristics showed no statistical difference in other basic data except the tumour-node-metastasis (pTNM) stage (p = 0.007). Univariate and multivariate analysis of survival risk factors showed that splenic characteristics (p = 0.042), age (p < 0.001), tumor location (p = 0.002), and pTNM stage (p < 0.001) were independent risk factors for survival. The prognostic prediction model combined with splenic characteristics significantly improved the accuracy of prognosis, predicting one-and three-year survival rates.

CONCLUSION

Splenic features extracted from imaging technology can accurately predict the long-term survival of patients with gastric cancer. Splenic characteristic grouping can effectively improve the accuracy of survival prediction and gastric cancer prognosis.