Artificial intelligence-assisted real-time monitoring of effective withdrawal time during colonoscopy: a novel quality marker of colonoscopy.

BACKGROUND AND AIMS: The importance of withdrawal time during colonoscopy cannot be overstated in mitigating the risk of missed lesions and postcolonoscopy colorectal cancer. We evaluated a novel colonoscopy quality metric called the effective withdrawal time (EWT), which is an artificial intelligence (AI)-derived quantitative measure of quality withdrawal time, and its association with various colonic lesion detection rates as compared with standard withdrawal time (SWT). METHODS: Three hundred fifty video recordings of colonoscopy withdrawal (from the cecum to the anus) were assessed by the new AI model. The primary outcome was adenoma detection rate (ADR) according to different quintiles of EWT. Multivariate logistic regression, adjusting for baseline covariates, was used to determine the adjusted odd ratios (ORs) for EWT on lesion detection rates, with the lowest quintile as reference. The area under the receiver-operating characteristic curve of EWT was compared with SWT. RESULTS: The crude ADR in different quintiles of EWT, from lowest to highest, was 10.0%, 31.4%, 33.3%, 53.5%, and 85.7%. The ORs of detecting adenomas and polyps were significantly higher in all top 4 quintiles when compared with the lowest quintile. Each minute increase in EWT was associated with a 49% increase in ADR (aOR, 1.49; 95% confidence interval [CI], 1.36-1.65). The area under the receiver-operating characteristic curve of EWT was also significantly higher than SWT on adenoma detection (.80 [95% CI, .75-.84] vs .70 [95% CI, .64-.74], P < .01). CONCLUSIONS: AI-derived monitoring of EWT is a promising novel quality indicator for colonoscopy, which is more associated with ADR than SWT.

Computer-assisted detection versus conventional colonoscopy for proximal colonic lesions: a multicenter, randomized, tandem-colonoscopy study.

BACKGROUND AND AIMS: Computer-assisted detection (CADe) is a promising technologic advance that enhances adenoma detection during colonoscopy. However, the role of CADe in reducing missed colonic lesions is uncertain. The aim of this study was to determine the miss rates of proximal colonic lesions by CADe and conventional colonoscopy. METHODS: This was a prospective, multicenter, randomized, tandem-colonoscopy study conducted in 3 Asian centers. Patients were randomized to receive CADe or conventional white-light colonoscopy during the first withdrawal of the proximal colon (cecum to splenic flexure), immediately followed by tandem examination of the proximal colon with white light in both groups. The primary outcome was adenoma/polyp miss rate, which was defined as any adenoma/polyp detected during the second examination. RESULTS: Of 223 patients (48.6% men; median age, 63 years) enrolled, 7 patients did not have tandem examination, leaving 108 patients in each group. There was no difference in the miss rate for proximal adenomas (CADe vs conventional: 20.0% vs 14.0%, P = .07) and polyps (26.7% vs 19.6%, P = .06). The CADe group, however, had significantly higher proximal polyp (58.0% vs 46.7%, P = .03) and adenoma (44.7% vs 34.6%, P = .04) detection rates than the conventional group. The mean number of proximal polyps and adenomas detected per patient during the first examination was also significantly higher in the CADe group (polyp: 1.20 vs .86, P = .03; adenoma, .91 vs .61, P = .03). Subgroup analysis showed that CADe enhanced proximal adenoma detection in patients with fair bowel preparation, shorter withdrawal time, and endoscopists with lower adenoma detection rate. CONCLUSIONS: This multicenter trial from Asia confirmed that CADe can further enhance proximal adenoma and polyp detection but may not be able to reduce the number of missed proximal colonic lesions. (Clinical trial registration number: NCT04294355.).

Impact of first wave of COVID-19 on outcomes of hospitalization for upper gastrointestinal bleeding in Hong Kong: a population-based study.

Background and study aims The COVID-19 pandemic has caused a major disruption in the healthcare system. This study determined the impact of the first wave of COVID-19 on the number and outcome of patients hospitalized for upper gastrointestinal bleeding (UGIB) in Hong Kong. Patients and methods Records of all patients hospitalized for UGIB in Hong Kong public hospitals between October 2018 and June 2020 were retrieved. The number and characteristics of patients hospitalized for UGIB after COVID-19 was compared by autoregressive integrated moving average (ARIMA) model prediction and historical cohort. Results Since the first local case of COVID-19, there was an initial drop in UGIB hospitalizations (observed 29.8 vs predicted 35.5 per week; P  = 0.05) followed by a rebound (39.8 vs 26.7 per week; P  < 0.01) with a turning point at week 14 (Petitt's test, P  < 0.001). There was a negative association between the number of COVID-19 cases and the number of patients hospitalized for UGIB (Pearson correlation -0.53, P  < 0.001). Patients admitted after the outbreak of COVID-19 had lower hemoglobin (7.5 vs baseline 8.3 g/dL; P  < 0.01) and a greater need for blood transfusion (64.5 % vs baseline 50.4 %; P  < 0.01), but similar rates of all-cause mortality (6.9 % vs 7.1 %; P  = 0.82) and rebleeding (6.7 % vs 5.1 %; P  = 0.11). There was also a higher proportion of patients with variceal bleeding (10.5 % vs baseline 5.3 %; P  < 0 .01). Conclusions There was a dynamic change in the number of patients hospitalized for UGIB in Hong Kong during the first wave of the COVID-19 outbreak, with more obvious impact during the initial phase only.

Applications of machine learning models in the prediction of gastric cancer risk in patients after Helicobacter pylori eradication.

BACKGROUND: The risk of gastric cancer after Helicobacter pylori (H. pylori) eradication remains unknown. AIM: To evaluate the performances of seven different machine learning models in predicting gastric cancer risk after H. pylori eradication. METHODS: We identified H. pylori-infected patients who had received clarithromycin-based triple therapy between 2003 and 2014 in Hong Kong. Patients were divided into training (n = 64 238) and validation sets (n = 25 330), according to period of eradication therapy. The data were used to construct seven machine learning models to predict risk of gastric cancer development within 5 years after H. pylori eradication. A total of 26 clinical variables were input into these models. The performances were measured by the area under receiver operating characteristic curve (AUC) analysis. RESULTS: During a mean follow-up of 4.7 years, 0.21% of H. pylori-eradicated patients developed gastric cancer. Of the seven machine learning models, extreme gradient boosting (XGBoost) had the best performance in predicting cancer development (AUC 0.97, 95%CI 0.96-0.98), and was superior to conventional logistic regression (AUC 0.90, 95% CI 0.84-0.92). With the XGBoost model, the number of patients considered at high risk of gastric cancer was 6.6%, with miss rate of 1.9%. Patient age, presence of intestinal metaplasia, and gastric ulcer were the heavily weighted factors used by the XGBoost. CONCLUSION: Based on simple baseline patient information, machine learning model can accurately predict the risk of post-eradication gastric cancer. This model could substantially reduce the number of patients who require endoscopic surveillance.

New insights on missed colonic lesions during colonoscopy through artificial intelligence-assisted real-time detection (with video).

BACKGROUND AND AIMS: Meta-analysis shows that up to 26% of adenomas could be missed during colonoscopy. We investigated whether the use of artificial intelligence (AI)-assisted real-time detection could provide new insights into mechanisms underlying missed lesions during colonoscopy. METHODS: A validated real-time deep-learning AI model for the detection of colonic polyps was first tested in videos of tandem colonoscopy of the proximal colon for missed lesions. The real-time AI model was then prospectively validated in a total colonoscopy in which the endoscopist was blinded to real-time AI findings. Segmental unblinding of the AI findings were provided, and the colonic segment was then re-examined when missed lesions were detected by AI but not the endoscopist. All polyps were removed for histologic examination as the criterion standard. RESULTS: Sixty-five videos of tandem examination of the proximal colon were reviewed by AI. AI detected 79.1% (19/24) of missed proximal adenomas in the video of the first-pass examination. In 52 prospective colonoscopies, real-time AI detection detected at least 1 missed adenoma in 14 patients (26.9%) and increased the total number of adenomas detected by 23.6%. Multivariable analysis showed that a missed adenoma(s) was more likely when there were multiple polyps (adjusted odds ratio, 1.05; 95% confidence interval, 1.02-1.09; P < .0001) or colonoscopy was performed by less-experienced endoscopists (adjusted odds ratio, 1.30; 95% confidence interval, 1.05-1.62; P = .02). CONCLUSIONS: Our findings provide new insights on the prominent role of human factors, including inexperience and distraction, on missed colonic lesions. With the use of real-time AI assistance, up to 80% of missed adenomas could be prevented. (Clinical trial registration number: NCT04227795.).

Is artificial intelligence the final answer to missed polyps in colonoscopy?

Lesions missed by colonoscopy are one of the main reasons for post-colonoscopy colorectal cancer, which is usually associated with a worse prognosis. Because the adenoma miss rate could be as high as 26%, it has been noted that endoscopists with higher adenoma detection rates are usually associated with lower adenoma miss rates. Artificial intelligence (AI), particularly the deep learning model, is a promising innovation in colonoscopy. Recent studies have shown that AI is not only accurate in colorectal polyp detection but can also reduce the miss rate. Nevertheless, the application of AI in real-time detection has been hindered by heterogeneity of the AI models and study design as well as a lack of long-term outcomes. Herein, we discussed the principle of various AI models and systematically reviewed the current data on the use of AI on colorectal polyp detection and miss rates. The limitations and future prospects of AI on colorectal polyp detection are also discussed.

Accuracy of artificial intelligence-assisted detection of upper GI lesions: a systematic review and meta-analysis.

BACKGROUND AND AIMS: Artificial intelligence (AI)-assisted detection is increasingly used in upper endoscopy. We performed a meta-analysis to determine the diagnostic accuracy of AI on detection of gastric and esophageal neoplastic lesions and Helicobacter pylori (HP) status. METHODS: We searched Embase, PubMed, Medline, Web of Science, and Cochrane databases for studies on AI detection of gastric or esophageal neoplastic lesions and HP status. After assessing study quality using the Quality Assessment of Diagnostic Accuracy Studies tool, a bivariate meta-analysis following a random-effects model was used to summarize the data and plot hierarchical summary receiver-operating characteristic curves. The diagnostic accuracy was determined by the area under the hierarchical summary receiver-operating characteristic curve (AUC). RESULTS: Twenty-three studies including 969,318 images were included. The AUC of AI detection of neoplastic lesions in the stomach, Barrett's esophagus, and squamous esophagus and HP status were .96 (95% confidence interval [CI], .94-.99), .96 (95% CI, .93-.99), .88 (95% CI, .82-.96), and .92 (95% CI, .88-.97), respectively. AI using narrow-band imaging was superior to white-light imaging on detection of neoplastic lesions in squamous esophagus (.92 vs .83, P < .001). The performance of AI was superior to endoscopists in the detection of neoplastic lesions in the stomach (AUC, .98 vs .87; P < .001), Barrett's esophagus (AUC, .96 vs .82; P < .001), and HP status (AUC, .90 vs .82; P < .001). CONCLUSIONS: AI is accurate in the detection of upper GI neoplastic lesions and HP infection status. However, most studies were based on retrospective reviews of selected images, which requires further validation in prospective trials.

Accuracy of artificial intelligence on histology prediction and detection of colorectal polyps: a systematic review and meta-analysis.

BACKGROUND AND AIMS: We performed a meta-analysis of all published studies to determine the diagnostic accuracy of artificial intelligence (AI) on histology prediction and detection of colorectal polyps. METHOD: We searched Embase, PubMed, Medline, Web of Science, and Cochrane library databases to identify studies using AI for colorectal polyp histology prediction and detection. The quality of included studies was measured by the Quality Assessment of Diagnostic Accuracy Studies tool. We used a bivariate meta-analysis following a random-effects model to summarize the data and plotted hierarchical summary receiver operating characteristic curves. The area under the hierarchical summary receiver operating characteristic curve (AUC) served as an indicator of the diagnostic accuracy and during head-to-head comparisons. RESULTS: A total of 7680 images of colorectal polyps from 18 studies were included in the analysis of histology prediction. The accuracy of the AI (AUC) was .96 (95% confidence interval [CI], .95-.98), with a corresponding pooled sensitivity of 92.3% (95% CI, 88.8%-94.9%) and specificity of 89.8% (95% CI, 85.3%-93.0%). The AUC of AI using narrow-band imaging (NBI) was significantly higher than the AUC using non-NBI (.98 vs .84, P < .01). The performance of AI was superior to nonexpert endoscopists (.97 vs .90, P < .01). For characterization of diminutive polyps using a deep learning model with nonmagnifying NBI, the pooled negative predictive value was 95.1% (95% CI, 87.7%-98.1%). For polyp detection, the pooled AUC was .90 (95% CI, .67-1.00) with a sensitivity of 95.0% (95% CI, 91.0%-97.0%) and a specificity of 88.0% (95% CI, 58.0%-99.0%). CONCLUSIONS: AI was accurate in histology prediction and detection of colorectal polyps, including diminutive polyps. The performance of AI was better under NBI and was superior to nonexpert endoscopists. Despite the difference in AI models and study designs, AI performances are rather consistent, which could serve as a reference for future AI studies.

Feedback from artificial intelligence improved the learning of junior endoscopists on histology prediction of gastric lesions.

Background and study aims Artificial intelligence (AI)-assisted image classification has been shown to have high accuracy on endoscopic diagnosis. We evaluated the potential effects of use of an AI-assisted image classifier on training of junior endoscopists for histological prediction of gastric lesions. Methods An AI image classifier was built on a convolutional neural network with five convolutional layers and three fully connected layers A Resnet backbone was trained by 2,000 non-magnified endoscopic gastric images. The independent validation set consisted of another 1,000 endoscopic images from 100 gastric lesions. The first part of the validation set was reviewed by six junior endoscopists and the prediction of AI was then disclosed to three of them (Group A) while the remaining three (Group B) were not provided this information. All endoscopists reviewed the second part of the validation set independently. Results The overall accuracy of AI was 91.0 % (95 % CI: 89.2-92.7 %) with 97.1 % sensitivity (95 % CI: 95.6-98.7%), 85.9 % specificity (95 % CI: 83.0-88.4 %) and 0.91 area under the ROC (AUROC) (95 % CI: 0.89-0.93). AI was superior to all junior endoscopists in accuracy and AUROC in both validation sets. The performance of Group A endoscopists but not Group B endoscopists improved on the second validation set (accuracy 69.3 % to 74.7 %; P  = 0.003). Conclusion The trained AI image classifier can accurately predict presence of neoplastic component of gastric lesions. Feedback from the AI image classifier can also hasten the learning curve of junior endoscopists in predicting histology of gastric lesions.

Linked color imaging versus narrow-band imaging for colorectal polyp detection: a prospective randomized tandem colonoscopy study.

BACKGROUND AND AIMS: Linked color imaging (LCI) is a newly available image-enhanced endoscopy (IEE) system that emphasizes the red mucosal color. No study has yet compared LCI with other available IEE systems. Our aim was to investigate polyp detection rates using LCI compared with narrow-band imaging (NBI). METHODS: This is a prospective randomized tandem colonoscopy study. Eligible patients who underwent colonoscopy for symptoms or screening/surveillance were randomized in a 1:1 ratio to receive tandem colonoscopy with both colonoscope withdrawals using LCI or NBI. The primary outcome was the polyp detection rate. RESULTS: Two hundred seventy-two patients were randomized (mean age, 62 years; 48.2% male; colonoscopy for symptoms, 72.8%) with 136 in each arm. During the first colonoscopy, the polyp detection rate (71.3% vs 55.9%; P = .008), serrated lesion detection rate (34.6% vs 22.1%; P = .02), and mean number of polyps detected (2.04 vs 1.35; P = .02) were significantly higher in the NBI group than in the LCI group. There was also a trend of higher adenoma detection rate in the NBI group compared with the LCI group (51.5% vs 39.7%, respectively; P = .05). Multivariable analysis confirmed that use of NBI (adjusted odds ratio, 1.99; 95% confidence interval, 1.09-3.68) and withdrawal time >8 minutes (adjusted odds ratio, 5.11; 95% confidence interval, 2.79-9.67) were associated with polyp detection. Overall, 20.5% of polyps and 18.1% of adenomas were missed by the first colonoscopy, but there was no significant difference in the miss rates between the 2 groups. CONCLUSION: NBI was significantly better than LCI for colorectal polyp detection. However, both LCI and NBI missed 20.5% of polyps. (Clinical trial registration number: NCT03336359.).

Endoscopic prediction of deeply submucosal invasive carcinoma with use of artificial intelligence.

Background and study aims We evaluated use of artificial intelligence (AI) assisted image classifier in determining the feasibility of curative endoscopic resection of large colonic lesion based on non-magnified endoscopic images Methods AI image classifier was trained by 8,000 endoscopic images of large (≥ 2 cm) colonic lesions. The independent validation set consisted of 567 endoscopic images from 76 colonic lesions. Histology of the resected specimens was used as gold standard. Curative endoscopic resection was defined as histology no more advanced than well-differentiated adenocarcinoma, ≤ 1 mm submucosal invasion and without lymphovascular invasion, whereas non-curative resection was defined as any lesion that could not meet the above requirements. Performance of the trained AI image classifier was compared with that of endoscopists. Results In predicting endoscopic curative resection, AI had an overall accuracy of 85.5 %. Images from narrow band imaging (NBI) had significantly higher accuracy (94.3 % vs 76.0 %; P  < 0.00001) and area under the ROC curve (AUROC) (0.934 vs 0.758; P  = 0.002) than images from white light imaging (WLI). AI was superior to two junior endoscopists in terms of accuracy (85.5 % vs 61.9 % or 82.0 %, P  < 0.05), AUROC (0.837 vs 0.638 or 0.717, P  < 0.05) and confidence level (90.1 % vs 83.7 % or 78.3 %, P  < 0.05). However, there was no statistical difference in accuracy and AUROC between AI and a senior endoscopist. Conclusions The trained AI image classifier based on non-magnified images can accurately predict probability of curative resection of large colonic lesions and is better than junior endoscopists. NBI images have better accuracy than WLI for AI prediction.

Territory-wide population-based study of chronic hepatitis C infection and implications for hepatitis elimination in Hong Kong.

BACKGROUND: To study the epidemiology of chronic hepatitis C virus infection in Hong Kong and to estimate the service gap for achieving the WHO hepatitis elimination targets of attaining a diagnosis rate of 90%, treatment rate of 80% and 65% reduction in mortality rate by 2030. METHODS: From January 2005 to March 2017, patients who were tested positive for anti-HCV were retrospectively retrieved from all public hospitals in Hong Kong. The epidemiological data of 15 participating hospitals were analysed. RESULTS: A total of 11 309 anti-HCV+ patients were identified and the estimated diagnosis rate was 50.9%. Our HCV-infected patients were ageing (median age 59). The all-cause mortality rate increased from 26.2 to 54.8 per 1000 person-years over the last decade. Our estimated treatment rate was 12.4%. Among the treated patients, 93.6% had received pegylated interferon/ribavirin (Peg-IFN/RBV) but only 10.8% had received interferon-free direct-acting antivirals (DAAs). In a cohort of 1533 patients, 39% already had advanced liver fibrosis or cirrhosis. The sustained virological response rate for Peg-IFN/RBV and DAAs were 74.8% and 97.2% respectively. However, more than 70% of patients were not subjected to interferon treatment for various reasons. Patients who achieved SVR were associated with a significantly lower risk of HCC (4.7% vs 9.6%, P = 0.005) and death (1.7% vs 23.8%, P < 0.001). CONCLUSION: Our diagnosis rate, treatment rate and mortality rate reduction were still low, particularly the Peg-IFN outcomes, making it difficult to meet the WHO hepatitis elimination targets. A more generalized use of DAAs is urgently needed to improve the situation.