Computed Tomography Texture Features and Risk Factor Analysis of Postoperative Recurrence of Patients with Advanced Gastric Cancer after Radical Treatment under Artificial Intelligence Algorithm

Abstract

Computer tomography texture analysis (CTTA) based on the V-Net convolutional neural network (CNN) algorithm was used to analyze the recurrence of advanced gastric cancer after radical treatment. Meanwhile, the clinical characteristics of patients were analyzed to explore the recurrence factors. 86 patients who underwent the advanced radical gastrectomy for gastric cancer were retrospectively selected as the research objects. Patients were divided into the no-recurrence group (30 cases) and the recurrence group (56 cases) according to whether there was recurrence after radical treatment. CTTA was performed before and after surgery in both groups to analyze the risk factors for recurrence. The results showed that the dice coefficient (0.9209) and the intersection over union (IOU) value (0.8392) of the V–CNN segmentation effect were signally higher than those of CNN, V-Net, and context encoder network (CE-Net) ( $P$  < 0.05). The mean value of arterial phase and portal phase (65.29 ± 9.23)/(79.89 ± 10.83), kurtosis (3.22)/(3.13), entropy (9.99 ± 0.53)/(9.97 ± 0.83), and correlation (4.12 × 10−5/4.21 × 10−5) of the recurrence group was higher than the no-recurrence group, while the skewness (0.01)/(−0.06) of the recurrence group was lower than that of the no-recurrence group ( $P$  < 0.05). Patients aged 60 years old and above, with a tumor diameter of 6 cm and above, and in the stage III/IV in the recurrence group were higher than those in the no-recurrence group, and patients with chemotherapy were lower ( $P$  < 0.05). To sum up, age, tumor diameter, whether chemotherapy should be performed, and tumor staging were all the risk factors of postoperative recurrence among patients with gastric cancer. Besides, CT texture parameter could be used to predict and analyze the postoperative recurrence of gastric cancer with good clinical application values.