Use Of Hisense CAS System And Deep Neural Networkenhances Infrapyloric Lymph Node Dissection For Gastric Cancer

[Background] The main process of radical gastrectomy is complete resection of the tumor and dissection of the perigastric lymph nodes(LNs).The extent of dissection and the minimum number of lymph nodes needed to be dissected have been the focus of controversy among surgeons.Currently,D2 dissection of perigastric LNs and dissection of at least 16 LNs has become a widely recognized standard.Perigastric LNs are responsible for the drainage from different parts of the stomach,for the most common gastric cancer in the lower part,no.6 group LN is the predilection siteof LNs metastasis,because it is in the infrapylorus area with the rich blood supply and intense organs.Many scholars committed to the research of no.6 group LNs,one of the most concern two aspects is the improvement of dissection method and its influence on prognosis of patients.The dissection of lymph nodes in the no.6 group is closely related to patient prognoses.The infrapyloric vessel-based method of LN dissection is accepted by most scholars,but there are many variations in the right gastroepiploic artery(RGEA),right gastroepiploic vein(RGEV),infrapyloric artery(IPA)and other blood vessels of patients,which can complicate the cleaning process.The purpose of this study was to establish a three-dimensional model of the infrapyloric vessels using a Hisense computer-assisted surgery(CAS)system before the operation,to understand the types of blood vessel variations that exist,and to discuss the metastasis status of LNs in the no.6 group.To use deep neural networks on computed tomography(CT)diagnosisof perigastric metastatic lymph nodes(PGMLNs)to simulate the radiologist's recognition of lymph nodes,to achieve more accurate identification results and to guide the choice of gastric cancer treatment plans and the evaluation of the prognosis.[Methods]One hundredand sixty patients with gastric cancer were included in the study.After screening with exclusion criteria,56 patients were not included in the study due to various reasons.Finally,104 patients with gastric cancer were randomly divided evenly betweena CAS group and a CT group,which underwentLN dissection guided by the Hisense CAS system and conventional LN dissection,respectively.In the CAS group,imaging simulation was performed before the operation to summarize the variations in the RGEA and RGEV as well as their spatial relationship,and used the three-dimensional model of the blood vessels and the intelligent gesture recognition system for surgical guidance during the operation.Intraoperative and postoperative complications in the two groups were recorded.The number of no.6 group LNs dissected and the metastasis status were compared between the two groups.Sequential deep learning sessions were performed with a faster region-based convolutional neural network(FR-CNN)on 60 enhanced CT images for each patient.A total of 1,123 images with suspected lymph node metastasis from enhanced abdominal CT scans of 293 patients with gastric cancer were identified and labeled by radiologists,which together with 17,580 original images were used for the FR-CNN deep learning.The identification results on 6,240 random CT images from 104 gastric cancer patients by the FR-CNN were compared with pathological examination in terms of their identification accuracy.[Results] In the 50 patients in the CAS group,the gastrocolic trunk of Henle(GTH)was mainly divided into a gastrocolic type(34.0%)and a gastropancreatic colonic type(66.0%);the RGEA was mainly divided into a crude blood supply type(24.0%)and a fine blood supply type(76.0%);and the relationship between the RGEA and RGEV was divided into an adjacent type(58.0%)and a separated type(42.0%).The time of LN dissection in the CAS group was significantly longer than that in the CT group,while the bleeding volume was similar between the two groups.Compared with the CT group,theCAS group had fewer cases of intraoperative gastrocolic trunk injury and postoperative pancreatic leakage.The occurrence of postoperative duodenal stump fistula and lymphatic leakage was similar between the groups.The total LN metastasis rate(LNMR)and LN metastasis degree(LNMD)in the no.6 group were 50.0% and 25.8%,respectively.Except in diffuse gastric cancer,the LNMR and LNMD were the highest for lower gastric cancer cases,at 55.9% and 27.6%,respectively.In the CAS group,no significant statistical difference was found in the comparison of LN metastasis in the no.6 group among different types of vascular variants.Among the groups of different tumor stages,the dissection and metastasis of no.6 group LNs were aggravated with the increase of tumor stages.There was a statistically significant difference in the number of no.6 group LNs dissected and metastasized in the CAS group compared with the CT group.The LNMR and LNMD of the CAS group were also higher than those of the CT group.Accordingto the multiple logistic regression model,tumor location and TNM stage were significantly correlatedwith group 6 LN metastases.The determination criteria for the metastatic LNs in the CT image identification process were short diameter,edge shape and enhanced density value of the enlarged LN,of which the most important quantitative parameter was the short diameter.In verifying the results of the two deep learning groups,their respective receiver operating characteristic(ROC)curves were generated,and the corresponding area under the curve(AUC)values were calculated as 0.9387.Among them,the accuracy rate of suspicious group 6 lymph nodes in CT identified by artificial intelligence was 95.1%.[Conclusion] By establishing a three-dimensional model of the infrapyloric vessels using a Hisense CAS system before the operation,we comprehensively determined the anatomical variations in each collateral vessel,which greatly assisted in the dissection of LNs during the operation.We also summarized the intraoperative and postoperative complications and the metastasis status of LNs in theno.6 group.The application of the Hisense CAS system significantly improved the number of LNs dissected and the discovery rate of LN metastases without increasing the incidence of intraoperative or postoperative complications.The application of a deep neural network in the CT diagnosis of metastatic lymph nodes of gastric cancer can largely replace experienced radiologists,with a higher diagnosis accuracy.The deep neural network approach has the advantages of fast diagnosis and simple operation,and it can accurately guide the choice of treatment plan for patients with gastric cancer.Intraoperatively,this approach provides regional information for the dissection of key lymph nodes while helping improve the imbalanced imaging resource distribution in Chinese medicine.