Identifying drug-drug interactions (DDIs) is essential to prevent adverse effects from polypharmacy. Although deep learning has advanced DDI identification, the gap between powerful models and their lack of clinical application and evaluation has hindered clinical benefits. Here, we developed a Multi-Dimensional Feature Fusion model named MDFF, which integrates one-dimensional simplified molecular input line entry system sequence features, two-dimensional molecular graph features, and three-dimensional geometric features to enhance drug representations for predicting DDIs. MDFF was trained and validated on two DDI datasets, evaluated across three distinct scenarios, and compared with advanced DDI prediction models using accuracy, precision, recall, area under the curve, and F1 score metrics. MDFF achieved state-of-the-art performance across all metrics. Ablation experiments showed that integrating multi-dimensional drug features yielded the best results. More importantly, we obtained adverse drug reaction reports uploaded by Xiangya Hospital of Central South University from 2021 to 2023 and used MDFF to identify potential adverse DDIs. Among 12 real-world adverse drug reaction reports, the predictions of 9 reports were supported by relevant evidence. Additionally, MDFF demonstrated the ability to explain adverse DDI mechanisms, providing insights into the mechanisms behind one specific report and highlighting its potential to assist practitioners in improving medical practice.

Objective:To develop an intelligent surveillance system for identifying upper gastrointestinal high-risk patients and assigning surveillance intervals, and to verify its efficacy.Methods:The endoscopic and pathological reports of 23 035 patients undergoing endoscopy at Renmin Hospital of Wuhan University from January to October 2021 were collected retrospectively. A training set of 17 934 patients (January to August) and a test set of 5 101 patients (September to October) were established. Keywords in the endoscopic and pathological reports were extracted by the intelligent surveillance system, and high-risk patients were automatically identified and classified into 7 risk levels. Then the standardized surveillance intervals were assigned based on the guideline. Guideline-based surveillance intervals assigned by expert endoscopists based on endoscopic and pathological reports were used as the golden standard. The accuracy of the intelligent surveillance system was calculated. Of the patients within the test set, 189 were hospitalized and the surveillance intervals given by physicians could be obtained from the electronic health records. The accuracy of the intelligent surveillance system with that of physicians from different departments was compared. Then 67 patients were randomly selected from 189 patients by simple random sampling to evaluate the adjunctive effect of the system in assigning surveillance intervals among 3 endoscopists.Results:The overall accuracy of the intelligent surveillance system in identifying upper gastrointestinal high-risk patients was 99.94% (5 098/5 101), and that of assigning surveillance intervals to correctly included patients was 100.00% (534/534). The intelligent surveillance system achieved significantly higher accuracy compared with all physicians from different departments [98.94% (187/189) VS 35.45% (67/189), χ2=118.01, P<0.001] as well as physicians from department of gastroenterology [100.00% (117/117) VS 24.79% (29/117), χ2=86.01, P<0.001]. With the assistance of the intelligent surveillance system, the endoscopists' accuracy of assigning surveillance intervals to 67 patients was significantly improved [55.22% (111/201) VS 22.39% (45/201), χ2=58.68, P<0.001]. Conclusion:The intelligent surveillance system can accurately identify upper gastrointestinal high-risk patients and assign surveillance intervals according to risk levels, which can alleviate the workload of doctors and improve the follow-up rate of patients.

Objective:To develop an artificial intelligence (AI) calculation system for the effective withdrawal time of colonoscopy and to evaluate its clinical application value.Methods:First, 17 118 colonoscopy pictures from Renmin Hospital of Wuhan University were used for training and testing to establish a deep convolutional neural network model to recognize various colonoscopy fields. Then this model was integrated with the internal and external recognition model and cecum recognition model developed by the research group to create an AI system for automatic calculation of the effective withdrawal time. Finally, 944 colonoscopy videos from the Endoscopy Center of Renmin Hospital of Wuhan University from July 1, 2020 to October 10, 2020 were included in a retrospective analysis. AI automatic computing system was used to calculate the effective withdrawal time, and 89 of them were manually calculated to evaluate the accuracy of the AI automatic computing system. The remaining 855 cases were divided into two groups according to AI calculations, namely, the effective withdrawal time <6 min group ( n=615) and the effective withdrawal time ≥6 min group ( n=240), and the differences in the overall detection rate of adenoma and polyp were compared and analyzed. Results:The accuracy of AI automatic calculation system for effective withdrawal time reached 92.1% (82/89). The overall adenoma detection rate in the group with effective withdrawal time ≥6 min was 37.5% (90/240), that in the group with effective withdrawal time <6 min was 19.0% (117/615), and the difference was statistically significant ( χ2=32.11, P<0.001). The overall polyp detection rate in the group with effective withdrawal time ≥6 min was 75.0% (180/240), and that in the group with effective withdrawal time <6 min was 45.2% (278/615), with statistical significance ( χ2=61.62, P<0.001). Conclusion:AI automatic computing system can accurately calculate the effective withdrawal time of colonoscopy, and can be used to monitor the effective withdrawal time of clinical colonoscopy. In addition, effective withdrawal time ≥6 min can effectively improve the detection rate of adenoma and polyps.

Objective:To evaluate the ability of ENDOANGEL artificial intelligence system to predict invasion depth and differentiation status of early gastric cancer using more diverse multi-center videos, and to test the performance of the new system upgraded from ENDOANGEL.Methods:Based on the completed 2020 man-machine competition for early gastric cancer diagnosis using single-center videos, the second man-machine competition was conducted in 2022, involving 30 endoscopists from 30 hospitals across 10 Chinese provinces. A multi-center video cohort was retrospectively collected from 12 institutions in 8 provinces/municipalities in China. The study proceeded in 3 stages. First, the ENDOANGEL was re-tested on multi-center videos, its performance on single and multi-center videos was compared, then the ENDOANGEL was upgraded to ENDOANGEL-2022. Second, the second man-machine competition was conducted between ENDOANGEL-2022 and 30 endoscopists using multi-center videos, and the performance between ENDOANGEL-2022, ENDOANGEL and endoscopists on multi-center videos were compared. Third, the ENDOANGEL-2022 was re-tested on the single-center videos previously collected in 2020, its performance on single and multi-center videos was also compared.Results:Compared with the performance on single-center videos, the sensitivity of ENDOANGEL for predicting submucosal invasion of early gastric cancer decreased significantly [18.18% (2/11) VS 70.00% (7/10), P=0.030], but demonstrated comparable ability to predict undifferentiated type of early gastric cancer ( P>0.05). On multi-center videos, in the respect of predicting submucosal invasion of early gastric cancer, the sensitivity of ENDOANGEL-2022 was higher than that of ENDOANGEL [40.00% (4/10) VS 18.18% (2/11), P=0.361], but inferior to that of 30 endoscopists [40.00% VS 52.04% (95% CI: 43.70%-60.38%), P<0.001]. The specificity of ENDOANGEL-2022 was lower than that of ENDOANGEL [82.86% (29/35) VS 100.00% (34/34), χ2=4.41, P=0.036] and higher than that of 30 endoscopists [82.86% VS 68.97% (95% CI: 60.83%-77.11%), P=0.018], the accuracy of ENDOANGEL-2022 was lower than that of ENDOANGEL [73.33% (33/45) VS 80.00% (36/45), χ2=0.56, P=0.455] and higher than that of 30 endoscopists [73.33% VS 65.30% (95% CI: 60.61%-69.99%), P=0.018]. In the respect of predicting undifferentiated type of early gastric cancer, the sensitivity of ENDOANGEL-2022 was higher than that of ENDOANGEL [71.43% (5/7) VS 57.14% (4/7), P>0.999] and 30 endoscopists [71.43% VS 63.11% (95% CI: 55.58%-70.64%), P=0.031], the specificity of ENDOANGEL-2022 was lower than that of ENDOANGEL [76.32% (29/38) VS 78.95% (30/38), χ2=0.08, P=0.783] and higher than that of 30 endoscopists [76.32% VS 65.27% (95% CI: 59.10%-71.44%), P=0.004],the accuracy of ENDOANGEL-2022 was similar to that of ENDOANGEL [75.56% (34/45) VS 75.56% (34/45), χ2=0.00, P>0.999] and higher than that of 30 endoscopists [75.56% VS 65.10% (95% CI: 59.96%- 70.24%), P<0.001]. Compared with performance in single center videos, the sensitivity [40.00% VS 60.00%(6/10), P=0.656], specificity [82.86% VS 93.75% (15/16), χ2=0.37, P=0.542] and accuracy [73.33% VS 80.77% (21/26), χ2=0.50, P=0.479] of ENDOANGEL-2022 for predicting submucosal invasion of early gastric cancer decreased; in predicting undifferentiated type of early gastric cancer, the sensitivity of ENDOANGEL-2022 increased [71.43% VS 37.50% (3/8), P=0.315], while the specificity [76.32% VS 100.00% (18/18), χ2=3.48, P=0.062] and accuracy [75.56% VS 80.77% (21/26), χ2=0.26, P=0.612] decreased. Conclusion:Multi-center cases introduce greater heterogeneity that may reduce artificial intelligence prediction accuracy, but the artificial intelligence system still outperforms endoscopists.

Objective:To explore status and influencing factors of surveillance in colorectal post-polypectomy patients.Methods:Patients who underwent colorectal polypectomy in Renmin Hospital of Wuhan University between April 1, 2019 and June 30, 2019 were retrospectively studied. The surveillance information was obtained through electronic health record and telephone call. Status and influencing factors of surveillance in colorectal post-polypectomy patients were evaluated. Logistic regression model was used for multivariate analysis to determine independent risk factors influencing surveillance.Results:A total of 268 colorectal post-polypectomy patients and their surveillance information were reviewed, of whom 153 (57.09%) patients received surveillance colonoscopy, and 115 (42.91%) patients did not. Univariate analysis showed that the source of patients (outpatients VS inpatients, χ 2=5.68, P=0.017), department (others VS department of gastroenterology, χ 2=6.64, P=0.010), and the number of polyps (1/(2~4)/≥5, χ2=7.32, P=0.026) influenced the outcome of surveillance. Logistic regression model indicated that department of gastroenterology ( P=0.039, OR=2.12, 95% CI:1.04-4.34), risk level 3 ( P=0.040, OR=1.92, 95% CI:1.03-3.58) and the number of polyps ≥5 ( P=0.016, OR=2.89, 95% CI:1.22-6.83) were independent risk factors influencing surveillance. Conclusion:Patients visit the department of gastroenterology or had a risk level 3 or ≥5 polyps are more likely to opt for surveillance following the procedure.

Objective:To analyze differences in detection rates for gastric precancerous lesions and neoplasms before versus after implementing a computer-aided detection (CADe) system in real-world clinical practice.Methods:Clinical data of patients who underwent gastroscopic examinations in two examination rooms (Room 1 and 2) in the Digestive Endoscopy Center of Renmin Hospital of Wuhan University were retrospectively collected during two periods: from January to June 2018 and from January to June 2021. Patients were stratified into four groups: CADe group (Room 2, 2021, using CADe), Pre-CADe group (Room 2, 2018, without CADe), 18-Con group (Room 1, 2018, without CADe), and 21-Con group (Room 1, 2021, without CADe). The differences in the detection rates of intestinal metaplasia and neoplasms between different groups were compared.Results:The detection rate of intestinal metaplasia in the CADe group was significantly higher than that in the Pre-CADe group [5.76% (198/3 437) VS 3.23% (100/3 092), χ2=23.856, P<0.001]. It was also significantly higher than that in the 21-Con group [5.76% (198/3 437) VS 2.73% (131/4 796), χ2=47.895, P<0.001]. The detection rate of neoplasms in the CADe group was significantly higher than that in the Pre-CADe group [3.23% (111/3 437) VS 1.58% (49/3 092), χ2=18.421, P<0.001] and the 21-Con group [3.23% (111/3 437) VS 1.79% (86/4 796), χ2=17.687, P<0.001]. Conclusion:The CADe system can significantly improve the detection rates of gastric intestinal metaplasia and neoplasms in clinical settings, potentially facilitating early diagnosis and treatment.

Identifying drug-drug interactions(DDIs)is essential to prevent adverse effects from polypharmacy.Although deep learning has advanced DDI identification,the gap between powerful models and their lack of clinical application and evaluation has hindered clinical benefits.Here,we developed a Multi-Dimensional Feature Fusion model named MDFF,which integrates one-dimensional simplified molec-ular input line entry system sequence features,two-dimensional molecular graph features,and three-dimensional geometric features to enhance drug representations for predicting DDIs.MDFF was trained and validated on two DDI datasets,evaluated across three distinct scenarios,and compared with advanced DDI prediction models using accuracy,precision,recall,area under the curve,and F1 score metrics.MDFF achieved state-of-the-art performance across all metrics.Ablation experiments showed that integrating multi-dimensional drug features yielded the best results.More importantly,we obtained adverse drug reaction reports uploaded by Xiangya Hospital of Central South University from 2021 to 2023 and used MDFF to identify potential adverse DDIs.Among 12 real-world adverse drug reaction reports,the predictions of 9 reports were supported by relevant evidence.Additionally,MDFF demon-strated the ability to explain adverse DDI mechanisms,providing insights into the mechanisms behind one specific report and highlighting its potential to assist practitioners in improving medical practice.

Objective:To develop an intelligent surveillance system for identifying upper gastrointestinal high-risk patients and assigning surveillance intervals, and to verify its efficacy.Methods:The endoscopic and pathological reports of 23 035 patients undergoing endoscopy at Renmin Hospital of Wuhan University from January to October 2021 were collected retrospectively. A training set of 17 934 patients (January to August) and a test set of 5 101 patients (September to October) were established. Keywords in the endoscopic and pathological reports were extracted by the intelligent surveillance system, and high-risk patients were automatically identified and classified into 7 risk levels. Then the standardized surveillance intervals were assigned based on the guideline. Guideline-based surveillance intervals assigned by expert endoscopists based on endoscopic and pathological reports were used as the golden standard. The accuracy of the intelligent surveillance system was calculated. Of the patients within the test set, 189 were hospitalized and the surveillance intervals given by physicians could be obtained from the electronic health records. The accuracy of the intelligent surveillance system with that of physicians from different departments was compared. Then 67 patients were randomly selected from 189 patients by simple random sampling to evaluate the adjunctive effect of the system in assigning surveillance intervals among 3 endoscopists.Results:The overall accuracy of the intelligent surveillance system in identifying upper gastrointestinal high-risk patients was 99.94% (5 098/5 101), and that of assigning surveillance intervals to correctly included patients was 100.00% (534/534). The intelligent surveillance system achieved significantly higher accuracy compared with all physicians from different departments [98.94% (187/189) VS 35.45% (67/189), χ2=118.01, P<0.001] as well as physicians from department of gastroenterology [100.00% (117/117) VS 24.79% (29/117), χ2=86.01, P<0.001]. With the assistance of the intelligent surveillance system, the endoscopists' accuracy of assigning surveillance intervals to 67 patients was significantly improved [55.22% (111/201) VS 22.39% (45/201), χ2=58.68, P<0.001]. Conclusion:The intelligent surveillance system can accurately identify upper gastrointestinal high-risk patients and assign surveillance intervals according to risk levels, which can alleviate the workload of doctors and improve the follow-up rate of patients.

Objective:To develop an artificial intelligence (AI) calculation system for the effective withdrawal time of colonoscopy and to evaluate its clinical application value.Methods:First, 17 118 colonoscopy pictures from Renmin Hospital of Wuhan University were used for training and testing to establish a deep convolutional neural network model to recognize various colonoscopy fields. Then this model was integrated with the internal and external recognition model and cecum recognition model developed by the research group to create an AI system for automatic calculation of the effective withdrawal time. Finally, 944 colonoscopy videos from the Endoscopy Center of Renmin Hospital of Wuhan University from July 1, 2020 to October 10, 2020 were included in a retrospective analysis. AI automatic computing system was used to calculate the effective withdrawal time, and 89 of them were manually calculated to evaluate the accuracy of the AI automatic computing system. The remaining 855 cases were divided into two groups according to AI calculations, namely, the effective withdrawal time <6 min group ( n=615) and the effective withdrawal time ≥6 min group ( n=240), and the differences in the overall detection rate of adenoma and polyp were compared and analyzed. Results:The accuracy of AI automatic calculation system for effective withdrawal time reached 92.1% (82/89). The overall adenoma detection rate in the group with effective withdrawal time ≥6 min was 37.5% (90/240), that in the group with effective withdrawal time <6 min was 19.0% (117/615), and the difference was statistically significant ( χ2=32.11, P<0.001). The overall polyp detection rate in the group with effective withdrawal time ≥6 min was 75.0% (180/240), and that in the group with effective withdrawal time <6 min was 45.2% (278/615), with statistical significance ( χ2=61.62, P<0.001). Conclusion:AI automatic computing system can accurately calculate the effective withdrawal time of colonoscopy, and can be used to monitor the effective withdrawal time of clinical colonoscopy. In addition, effective withdrawal time ≥6 min can effectively improve the detection rate of adenoma and polyps.

Objective:To evaluate the ability of ENDOANGEL artificial intelligence system to predict invasion depth and differentiation status of early gastric cancer using more diverse multi-center videos, and to test the performance of the new system upgraded from ENDOANGEL.Methods:Based on the completed 2020 man-machine competition for early gastric cancer diagnosis using single-center videos, the second man-machine competition was conducted in 2022, involving 30 endoscopists from 30 hospitals across 10 Chinese provinces. A multi-center video cohort was retrospectively collected from 12 institutions in 8 provinces/municipalities in China. The study proceeded in 3 stages. First, the ENDOANGEL was re-tested on multi-center videos, its performance on single and multi-center videos was compared, then the ENDOANGEL was upgraded to ENDOANGEL-2022. Second, the second man-machine competition was conducted between ENDOANGEL-2022 and 30 endoscopists using multi-center videos, and the performance between ENDOANGEL-2022, ENDOANGEL and endoscopists on multi-center videos were compared. Third, the ENDOANGEL-2022 was re-tested on the single-center videos previously collected in 2020, its performance on single and multi-center videos was also compared.Results:Compared with the performance on single-center videos, the sensitivity of ENDOANGEL for predicting submucosal invasion of early gastric cancer decreased significantly [18.18% (2/11) VS 70.00% (7/10), P=0.030], but demonstrated comparable ability to predict undifferentiated type of early gastric cancer ( P>0.05). On multi-center videos, in the respect of predicting submucosal invasion of early gastric cancer, the sensitivity of ENDOANGEL-2022 was higher than that of ENDOANGEL [40.00% (4/10) VS 18.18% (2/11), P=0.361], but inferior to that of 30 endoscopists [40.00% VS 52.04% (95% CI: 43.70%-60.38%), P<0.001]. The specificity of ENDOANGEL-2022 was lower than that of ENDOANGEL [82.86% (29/35) VS 100.00% (34/34), χ2=4.41, P=0.036] and higher than that of 30 endoscopists [82.86% VS 68.97% (95% CI: 60.83%-77.11%), P=0.018], the accuracy of ENDOANGEL-2022 was lower than that of ENDOANGEL [73.33% (33/45) VS 80.00% (36/45), χ2=0.56, P=0.455] and higher than that of 30 endoscopists [73.33% VS 65.30% (95% CI: 60.61%-69.99%), P=0.018]. In the respect of predicting undifferentiated type of early gastric cancer, the sensitivity of ENDOANGEL-2022 was higher than that of ENDOANGEL [71.43% (5/7) VS 57.14% (4/7), P>0.999] and 30 endoscopists [71.43% VS 63.11% (95% CI: 55.58%-70.64%), P=0.031], the specificity of ENDOANGEL-2022 was lower than that of ENDOANGEL [76.32% (29/38) VS 78.95% (30/38), χ2=0.08, P=0.783] and higher than that of 30 endoscopists [76.32% VS 65.27% (95% CI: 59.10%-71.44%), P=0.004],the accuracy of ENDOANGEL-2022 was similar to that of ENDOANGEL [75.56% (34/45) VS 75.56% (34/45), χ2=0.00, P>0.999] and higher than that of 30 endoscopists [75.56% VS 65.10% (95% CI: 59.96%- 70.24%), P<0.001]. Compared with performance in single center videos, the sensitivity [40.00% VS 60.00%(6/10), P=0.656], specificity [82.86% VS 93.75% (15/16), χ2=0.37, P=0.542] and accuracy [73.33% VS 80.77% (21/26), χ2=0.50, P=0.479] of ENDOANGEL-2022 for predicting submucosal invasion of early gastric cancer decreased; in predicting undifferentiated type of early gastric cancer, the sensitivity of ENDOANGEL-2022 increased [71.43% VS 37.50% (3/8), P=0.315], while the specificity [76.32% VS 100.00% (18/18), χ2=3.48, P=0.062] and accuracy [75.56% VS 80.77% (21/26), χ2=0.26, P=0.612] decreased. Conclusion:Multi-center cases introduce greater heterogeneity that may reduce artificial intelligence prediction accuracy, but the artificial intelligence system still outperforms endoscopists.