Published online Jun 28, 2022. doi: 10.3748/wjg.v28.i24.2721
Peer-review started: October 30, 2021
First decision: March 11, 2022
Revised: March 25, 2022
Accepted: May 8, 2022
Article in press: May 8, 2022
Published online: June 28, 2022
Processing time: 237 Days and 2.6 Hours
With the increasing rate of diagnosis at early stages of gastric cancer, endoscopic submucosal dissection (ESD) is being actively applied as the minimally invasive treatment. Bleeding is one of the significant complications, with an incidence of 3.6%–6.9%. Because bleeding after ESD requires hospitalization and hemostatic interventions, there is a need to predict patients at a high risk of bleeding after ESD.
Currently, artificial intelligence systems are being applied in various fields of gastroenterology. Deep learning among artificial intelligence systems was automatically trained so that it could be generalized well. There has been no study on the efficacy of deep learning for predicting post-ESD bleeding (PEB), and no study has compared these systems with a clinical model.
This study aimed to develop and compare the performance of the deep learning and clinical model for predicting PEB in early gastric cancer (EGC) patients.
Patients who underwent ESD for EGC between January 2010 and June 2020 at the Samsung Medical Center, Seoul, South Korea, were screened retrospectively. We built a deep learning model and a clinical model based on the development set, which comprised 80% of the overall cohort. Subsequently, we validated the deep learning and clinical models in the validation set, which comprised 20% of the overall cohort. The deep learning model and the clinical model for prediction of bleeding after ESD were evaluated using two methods. First, sensitivity, specificity, positive predictive value, negative predictive value, and receiver operating characteristic area curve along with the area under the curve (AUC) were analyzed. The performance with AUC was compared using the bootstrap test. Second, the risk stratification of PEB based on the development set was applied to the validation set and compared with the actual bleeding rate. The authors selected cutoff to discriminate the risk categories as low-, intermediate-, and high-risk at a bleeding rate of < 5% and < 9% in the development set referred to in a previous report.
Of the 5629 patients, 325 experienced PEB. The AUC for predicting PEB was 0.71 (95% confidence interval: 0.63-0.78) in the deep learning model and 0.70 (95% confidence interval: 0.62-0.77) in the clinical model, without significant difference (P = 0.730). In the validated set, the deep learning model showed an actual bleeding rate of 2.2%, 3.9%, and 11.6% in low-, intermediate-, high-risk categories, respectively; the clinical model showed an actual bleeding rate of 4.0%, 8.8%, and 18.2% in low-, intermediate-, high-risk categories, respectively.
In conclusion, we introduced a deep learning model to predict the risk of bleeding after ESD in patients with EGC. The model demonstrated its performance as comparable to the clinical model.
Based on the risk-prediction model, physicians could carefully assess the bleeding risk and perform preventive hemostasis during the procedure. Suppose additional management like the shielding method for preventing PEB in the selected high-risk group is attempted; in that case, it is anticipated that the deep learning model could support risk stratification.