Validation of artificial intelligence-based bowel preparation assessment in screening colonoscopy (with video).

BACKGROUND AND AIMS: Accurate bowel preparation assessment is essential for determining colonoscopy screening intervals. Patients with suboptimal bowel preparation are at a high risk of missing >5 mm adenomas and should undergo an early repeat colonoscopy. In this study, we used artificial intelligence (AI) to evaluate bowel preparation and validated the ability of the system to accurately identify patients who are at high risk of having >5 mm adenomas missed due to inadequate bowel preparation. METHODS: This prospective, single-center, observational study was conducted at the Eighth Affiliated Hospital, Sun Yat-sen University, from October 8, 2021, to November 9, 2022. Eligible patients who underwent screening colonoscopy were consecutively enrolled. The AI assessed bowel preparation using the e-Boston Bowel Preparation Scale (e-BBPS) while endoscopists made evaluations using BBPS. If both BBPS and e-BBPS deemed preparation adequate, the patient immediately underwent a second colonoscopy; otherwise, the patient underwent bowel re-cleansing before the second colonoscopy. RESULTS: Among the 393 patients, 72 adenomas >5 mm in size were detected; 27 adenomas >5 mm in size were missed. In unqualified-AI patients, the >5 mm adenoma miss rate (AMR) was significantly higher than in qualified-AI patients (35.71% vs 13.19% [P = .0056]; odds ratio [OR], .2734 [95% CI, .1139-.6565]), as were the AMR (50.89% vs 20.79% [P < .001]; OR, .2532 [95% CI, .1583-.4052]) and >5 mm polyp miss rate (35.82% vs 19.48% [P = .0152]; OR, .4335 [95% CI, .2288-.8213]). CONCLUSIONS: This study confirmed that patients classified as inadequate by AI exhibited an unacceptable >5 mm AMR, providing key evidence for implementing AI in guiding bowel re-cleansing and potentially standardizing future colonoscopy screening. (Clinical trial registration number: NCT05145712.).

Effect of artificial intelligence on novice-performed colonoscopy: a multicenter randomized controlled tandem study.

BACKGROUND AND AIMS: The efficacy and safety of colonoscopy performed by artificial intelligence (AI)-assisted novices remain unknown. The aim of this study was to compare the lesion detection capability of novices, AI-assisted novices, and experts. METHODS: This multicenter, randomized, noninferiority tandem study was conducted across 3 hospitals in China from May 1, 2022, to November 11, 2022. Eligible patients were randomized into 1 of 3 groups: the CN group (control novice group, withdrawal performed by a novice independently), the AN group (AI-assisted novice group, withdrawal performed by a novice with AI assistance), or the CE group (control expert group, withdrawal performed by an expert independently). Participants underwent a repeat colonoscopy conducted by an AI-assisted expert to evaluate the lesion miss rate and ensure lesion detection. The primary outcome was the adenoma miss rate (AMR). RESULTS: A total of 685 eligible patients were analyzed: 229 in the CN group, 227 in the AN group, and 229 in the CE group. Both AMR and polyp miss rate were lower in the AN group than in the CN group (18.82% vs 43.69% [P < .001] and 21.23% vs 35.38% [P < .001], respectively). The noninferiority margin was met between the AN and CE groups of both AMR and polyp miss rate (18.82% vs 26.97% [P = .202] and 21.23% vs 24.10% [P < .249]). CONCLUSIONS: AI-assisted colonoscopy lowered the AMR of novices, making them noninferior to experts. The withdrawal technique of new endoscopists can be enhanced by AI-assisted colonoscopy. (Clinical trial registration number: NCT05323279.).

A real-time deep learning-based system for colorectal polyp size estimation by white-light endoscopy: development and multicenter prospective validation.

BACKGROUND: The choice of polypectomy device and surveillance intervals for colorectal polyps are primarily decided by polyp size. We developed a deep learning-based system (ENDOANGEL-CPS) to estimate colorectal polyp size in real time. METHODS: ENDOANGEL-CPS calculates polyp size by estimating the distance from the endoscope lens to the polyp using the parameters of the lens. The depth estimator network was developed on 7297 images from five virtually produced colon videos and tested on 730 images from seven virtual colon videos. The performance of the system was first evaluated in nine videos of a simulated colon with polyps attached, then tested in 157 real-world prospective videos from three hospitals, with the outcomes compared with that of nine endoscopists over 69 videos. Inappropriate surveillance recommendations caused by incorrect estimation of polyp size were also analyzed. RESULTS: The relative error of depth estimation was 11.3% (SD 6.0%) in successive virtual colon images. The concordance correlation coefficients (CCCs) between system estimation and ground truth were 0.89 and 0.93 in images of a simulated colon and multicenter videos of 157 polyps. The mean CCC of ENDOANGEL-CPS surpassed all endoscopists (0.89 vs. 0.41 [SD 0.29]; P<0.001). The relative accuracy of ENDOANGEL-CPS was significantly higher than that of endoscopists (89.9% vs. 54.7%; P<0.001). Regarding inappropriate surveillance recommendations, the system's error rate is also lower than that of endoscopists (1.5% vs. 16.6%; P<0.001). CONCLUSIONS: ENDOANGEL-CPS could potentially improve the accuracy of colorectal polyp size measurements and size-based surveillance intervals.

Assessment of the role of false-positive alerts in computer-aided polyp detection for assistance capabilities.

BACKGROUND AND AIM: False positives (FPs) pose a significant challenge in the application of artificial intelligence (AI) for polyp detection during colonoscopy. The study aimed to quantitatively evaluate the impact of computer-aided polyp detection (CADe) systems' FPs on endoscopists. METHODS: The model's FPs were categorized into four gradients: 0-5, 5-10, 10-15, and 15-20 FPs per minute (FPPM). Fifty-six colonoscopy videos were collected for a crossover study involving 10 endoscopists. Polyp missed rate (PMR) was set as primary outcome. Subsequently, to further verify the impact of FPPM on the assistance capability of AI in clinical environments, a secondary analysis was conducted on a prospective randomized controlled trial (RCT) from Renmin Hospital of Wuhan University in China from July 1 to October 15, 2020, with the adenoma detection rate (ADR) as primary outcome. RESULTS: Compared with routine group, CADe reduced PMR when FPPM was less than 5. However, with the continuous increase of FPPM, the beneficial effect of CADe gradually weakens. For secondary analysis of RCT, a total of 956 patients were enrolled. In AI-assisted group, ADR is higher when FPPM ≤ 5 compared with FPPM > 5 (CADe group: 27.78% vs 11.90%; P = 0.014; odds ratio [OR], 0.351; 95% confidence interval [CI], 0.152-0.812; COMBO group: 38.40% vs 23.46%, P = 0.029; OR, 0.427; 95% CI, 0.199-0.916). After AI intervention, ADR increased when FPPM ≤ 5 (27.78% vs 14.76%; P = 0.001; OR, 0.399; 95% CI, 0.231-0.690), but no statistically significant difference was found when FPPM > 5 (11.90% vs 14.76%, P = 0.788; OR, 1.111; 95% CI, 0.514-2.403). CONCLUSION: The level of FPs of CADe does affect its effectiveness as an aid to endoscopists, with its best effect when FPPM is less than 5.

The effect of incorporating domain knowledge with deep learning in identifying benign and malignant gastric whitish lesions: A retrospective study.

BACKGROUND AND AIM: Early whitish gastric neoplasms can be easily misdiagnosed; differential diagnosis of gastric whitish lesions remains a challenge. We aim to build a deep learning (DL) model to diagnose whitish gastric neoplasms and explore the effect of adding domain knowledge in model construction. METHODS: We collected 4558 images from two institutions to train and test models. We first developed two sole DL models (1 and 2) using supervised and semi-supervised algorithms. Then we selected diagnosis-related features through literature research and developed feature-extraction models to determine features including boundary, surface, roundness, depression, and location. Then predictions of the five feature-extraction models and sole DL model were combined and inputted into seven machine-learning (ML) based fitting-diagnosis models. The optimal model was selected as ENDOANGEL-WD (whitish-diagnosis) and compared with endoscopists. RESULTS: Sole DL 2 had higher sensitivity (83.12% vs 68.67%, Bonferroni adjusted P = 0.024) than sole DL 1. Adding domain knowledge, the decision tree performed best among the seven ML models, achieving higher specificity than DL 1 (84.38% vs 72.27%, Bonferroni adjusted P < 0.05) and higher accuracy than DL 2 (80.47%, Bonferroni adjusted P < 0.001) and was selected as ENDOANGEL-WD. ENDOANGEL-WD showed better accuracy compared with 10 endoscopists (75.70%, P < 0.001). CONCLUSIONS: We developed a novel system ENDOANGEL-WD combining domain knowledge and traditional DL to detect gastric whitish neoplasms. Adding domain knowledge improved the performance of traditional DL, which provided a novel solution for establishing diagnostic models for other rare diseases potentially.

Current status of quality control in screening esophagogastroduodenoscopy and the emerging role of artificial intelligence.

Esophagogastroduodenoscopy (EGD) screening is being implemented in countries with a high incidence of upper gastrointestinal (UGI) cancer. High-quality EGD screening ensures the yield of early diagnosis and prevents suffering from advanced UGI cancer and minimal operational-related discomfort. However, performance varied dramatically among endoscopists, and quality control for EGD screening remains suboptimal. Guidelines have recommended potential measures for endoscopy quality improvement and research has been conducted for evidence. Moreover, artificial intelligence offers a promising solution for computer-aided diagnosis and quality control during EGD examinations. In this review, we summarized the key points for quality assurance in EGD screening based on current guidelines and evidence. We also outline the latest evidence, limitations, and future prospects of the emerging role of artificial intelligence in EGD quality control, aiming to provide a foundation for improving the quality of EGD screening.

Effect of a deep learning-based automatic upper GI endoscopic reporting system: a randomized crossover study (with video).

BACKGROUND AND AIMS: EGD is essential for GI disorders, and reports are pivotal to facilitating postprocedure diagnosis and treatment. Manual report generation lacks sufficient quality and is labor intensive. We reported and validated an artificial intelligence-based endoscopy automatic reporting system (AI-EARS). METHODS: The AI-EARS was designed for automatic report generation, including real-time image capturing, diagnosis, and textual description. It was developed using multicenter datasets from 8 hospitals in China, including 252,111 images for training, 62,706 images, and 950 videos for testing. Twelve endoscopists and 44 endoscopy procedures were consecutively enrolled to evaluate the effect of the AI-EARS in a multireader, multicase, crossover study. The precision and completeness of the reports were compared between endoscopists using the AI-EARS and conventional reporting systems. RESULTS: In video validation, the AI-EARS achieved completeness of 98.59% and 99.69% for esophageal and gastric abnormality records, respectively, accuracies of 87.99% and 88.85% for esophageal and gastric lesion location records, and 73.14% and 85.24% for diagnosis. Compared with the conventional reporting systems, the AI-EARS achieved greater completeness (79.03% vs 51.86%, P < .001) and accuracy (64.47% vs 42.81%, P < .001) of the textual description and completeness of the photo-documents of landmarks (92.23% vs 73.69%, P < .001). The mean reporting time for an individual lesion was significantly reduced (80.13 ± 16.12 seconds vs 46.47 ± 11.68 seconds, P < .001) after the AI-EARS assistance. CONCLUSIONS: The AI-EARS showed its efficacy in improving the accuracy and completeness of EGD reports. It might facilitate the generation of complete endoscopy reports and postendoscopy patient management. (Clinical trial registration number: NCT05479253.).

Effect of an automated flexible endoscope channel brushing system on improving reprocessing quality: a randomized controlled study.

BACKGROUND: Qualified reprocessing, of which meticulous channel brushing is the most crucial step, is essential for prevention and control of endoscopy-associated infections. However, channel brushing is often omitted in practice. This study aimed to evaluate the effect of an automated flexible endoscope channel brushing system (AECBS) on improving the quality of endoscope reprocessing. METHODS: This prospective, randomized controlled study was conducted between 24 November 2021 and 22 January 2022 at Renmin Hospital of Wuhan University, China. Eligible endoscopes were randomly allocated to the auto group (channels brushed by AECBS) or the manual group (channels brushed manually), with sampling and culturing after high-level disinfection and drying. The primary end point was the proportion of endoscopes with positive cultures. RESULTS: 204 endoscopes in the auto group and 205 in the manual group were analyzed. The proportion of endoscopes with positive cultures was significantly lower in the auto group (15.2 % [95 %CI 10.7 %-21.0 %]) than in the manual group (23.4 % [95 %CI 17.9 %-29.9 %]). CONCLUSIONS: AECBS could effectively reduce bioburden and improve reprocessing quality of gastroscopes and colonoscopes. AECBS has the potential to replace manual brushing and lower the risk of endoscopy-associated infections, providing a new option for the optimization of reprocessing.

Development and validation of an artificial intelligence-based system for predicting colorectal cancer invasion depth using multi-modal data.

OBJECTIVES: Accurate endoscopic optical prediction of the depth of cancer invasion is critical for guiding an optimal treatment approach of large sessile colorectal polyps but was hindered by insufficient endoscopists expertise and inter-observer variability. We aimed to construct a clinically applicable artificial intelligence (AI) system for the identification of presence of cancer invasion in large sessile colorectal polyps. METHODS: A deep learning-based colorectal cancer invasion calculation (CCIC) system was constructed. Multi-modal data including clinical information, white light (WL) and image-enhanced endoscopy (IEE) were included for training. The system was trained using 339 lesions and tested on 198 lesions across three hospitals. Man-machine contest, reader study and video validation were further conducted to evaluate the performance of CCIC. RESULTS: The overall accuracy of CCIC system using image and video validation was 90.4% and 89.7%, respectively. In comparison with 14 endoscopists, the accuracy of CCIC was comparable with expert endoscopists but superior to all the participating senior and junior endoscopists in both image and video validation set. With CCIC augmentation, the average accuracy of junior endoscopists improved significantly from 75.4% to 85.3% (P = 0.002). CONCLUSIONS: This deep learning-based CCIC system may play an important role in predicting the depth of cancer invasion in colorectal polyps, thus determining treatment strategies for these large sessile colorectal polyps.

Real-time automated diagnosis of colorectal cancer invasion depth using a deep learning model with multimodal data (with video).

BACKGROUND AND AIMS: The optical diagnosis of colorectal cancer (CRC) invasion depth with white light (WL) and image-enhanced endoscopy (IEE) remains challenging. We aimed to construct and validate a 2-modal deep learning-based system, incorporated with both WL and IEE images (named Endo-CRC) in estimating the invasion depth of CRC. METHODS: Samples were retrospectively obtained from 3 hospitals in China. We combined WL and IEE images into image pairs. Altogether, 337,278 image pairs from 268 noninvasive and superficial CRC and 181,934 image pairs from 82 deep CRC were used for training. A total of 296,644 and 4528 image pairs were used for internal and external tests and for comparison with endoscopists. Thirty-five videos were used for evaluating the real-time performance of the Endo-CRC system. Two deep learning models, solely using either WL (model W) or IEE images (model I), were constructed to compare with Endo-CRC. RESULTS: The accuracies of Endo-CRC in internal image tests with and without advanced CRC were 91.61% and 93.78%, respectively, and 88.65% in the external test, which did not include advanced CRC. In an endoscopist-machine competition, Endo-CRC achieved an expert comparable accuracy of 88.11% and the highest sensitivity compared with all endoscopists. In a video test, Endo-CRC achieved an accuracy of 100.00%. Compared with model W and model I, Endo-CRC had a higher accuracy (per image pair: 91.61% vs 88.27% compared with model I and 91.61% vs 81.32% compared with model W). CONCLUSIONS: The Endo-CRC system has great potential for assisting in CRC invasion depth diagnosis and may be well applied in clinical practice.

Effect of an artificial intelligence-based quality improvement system on efficacy of a computer-aided detection system in colonoscopy: a four-group parallel study.

BACKGROUND: Tandem colonoscopy studies have found that about one in five adenomas are missed at colonoscopy. It remains debatable whether the combination of a computer-aided polyp detection (CADe) system with a computer-aided quality improvement (CAQ) system for real-time monitoring of withdrawal speed results in additional benefits in adenoma detection or if the synergetic effect may be harmed due to excessive visual burden resulting from information overload. This study aimed to evaluate the interaction effect on improving the adenoma detection rate (ADR). METHODS: This single-center, randomized, four-group, parallel, controlled study was performed at Renmin Hospital of Wuhan University. Between 1 July and 15 October 2020, 1076 patients were randomly allocated into four treatment groups: control 271, CADe 268, CAQ 269, and CADe plus CAQ (COMBO) 268. The primary outcome was ADR. RESULTS: The ADR in the control, CADe, CAQ, and COMBO groups was 14.76 % (95 % confidence interval [CI] 10.54 to 18.98), 21.27 % (95 %CI 16.37 to 26.17), 24.54 % (95 %CI 19.39 to 29.68), and 30.60 % (95 %CI 25.08 to 36.11), respectively. The ADR was higher in the COMBO group compared with the CADe group (21.27 % vs. 30.6 %, P = 0.024, odds ratio [OR] 1.284, 95 %CI 1.033 to 1.596) but not compared with the CAQ group (24.54 % vs. 30.6 %, P = 0.213, OR 1.309, 95 %CI 0.857 to 2.000, respectively). CONCLUSIONS: CAQ significantly improved the efficacy of CADe in a four-group, parallel, controlled study. No significant difference in the ADR or polyp detection rate was found between CAQ and COMBO.

Evaluation of the effects of an artificial intelligence system on endoscopy quality and preliminary testing of its performance in detecting early gastric cancer: a randomized controlled trial.

BACKGROUND: Esophagogastroduodenoscopy (EGD) is a prerequisite for detecting upper gastrointestinal lesions especially early gastric cancer (EGC). An artificial intelligence system has been shown to monitor blind spots during EGD. In this study, we updated the system (ENDOANGEL), verified its effectiveness in improving endoscopy quality, and pretested its performance in detecting EGC in a multicenter randomized controlled trial. METHODS: ENDOANGEL was developed using deep convolutional neural networks and deep reinforcement learning. Patients undergoing EGD in five hospitals were randomly assigned to the ENDOANGEL-assisted group or to a control group without use of ENDOANGEL. The primary outcome was the number of blind spots. Secondary outcomes included performance of ENDOANGEL in predicting EGC in a clinical setting. RESULTS: 1050 patients were randomized, and 498 and 504 patients in the ENDOANGEL and control groups, respectively, were analyzed. Compared with the control group, the ENDOANGEL group had fewer blind spots (mean 5.38 [standard deviation (SD) 4.32] vs. 9.82 [SD 4.98]; P < 0.001) and longer inspection time (5.40 [SD 3.82] vs. 4.38 [SD 3.91] minutes; P < 0.001). In the ENDOANGEL group, 196 gastric lesions with pathological results were identified. ENDOANGEL correctly predicted all three EGCs (one mucosal carcinoma and two high grade neoplasias) and two advanced gastric cancers, with a per-lesion accuracy of 84.7 %, sensitivity of 100 %, and specificity of 84.3 % for detecting gastric cancer. CONCLUSIONS: In this multicenter study, ENDOANGEL was an effective and robust system to improve the quality of EGD and has the potential to detect EGC in real time.

Anti-Inflammatory Effects of Mytilus coruscus Polysaccharide on RAW264.7 Cells and DSS-Induced Colitis in Mice.

Considerable literature has been published on polysaccharides, which play a critical role in regulating the pathogenesis of inflammation and immunity. In this essay, the anti-inflammatory effect of Mytilus coruscus polysaccharide (MP) on lipopolysaccharide-stimulated RAW264.7 cells and a dextran sulfate sodium (DSS)-induced ulcerative colitis model in mice was investigated. The results showed that MP effectively promoted the proliferation of RAW264.7 cells, ameliorated the excessive production of inflammatory cytokines (TNF-α, IL-6, and IL-10), and inhibited the activation of the NF-κB signaling pathway. For DSS-induced colitis in mice, MP can improve the clinical symptoms of colitis, inhibit the weight loss of mice, reduce the disease activity index, and have a positive effect on the shortening of the colon caused by DSS, meliorating intestinal barrier integrity and lowering inflammatory cytokines in serum. Moreover, MP makes a notable contribution to the richness and diversity of the intestinal microbial community, and also regulates the structural composition of the intestinal flora. Specifically, mice treated with MP showed a repaired Firmicutes/Bacteroidetes ratio and an increased abundance of some probiotics like Anaerotruncus, Lactobacillus, Desulfovibrio, Alistipe, Odoribacter, and Enterorhabdus in colon. These data suggest that the MP could be a promising dietary candidate for enhancing immunity and protecting against ulcerative colitis.

Deep learning-based pancreas segmentation and station recognition system in EUS: development and validation of a useful training tool (with video).

BACKGROUND AND AIMS: EUS is considered one of the most sensitive modalities for pancreatic cancer detection, but it is highly operator-dependent and the learning curve is steep. In this study, we constructed a system named BP MASTER (pancreaticobiliary master) for EUS training and quality control. METHODS: The standard procedure of pancreatic EUS was divided into 6 stations. We developed a station classification model and a pancreas/abdominal aorta/portal confluence segmentation model with 19,486 images and 2207 images, respectively. Then, we used 1920 images and 700 images for classification and segmentation internal validation, respectively. To test station recognition we used 396 videos clips. An independent data set containing 180 images was applied for comparing the performance between models and EUS experts. Seven hundred sixty-eight images from 2 other hospitals were used for external validation. A crossover study was conducted to test the system effect on reducing difficulty in ultrasonographics interpretation among trainees. RESULTS: The models achieved 94.2% accuracy in station classification and .836 dice in segmentation at internal validation. At external validation, the models achieved 82.4% accuracy in station classification and .715 dice in segmentation. For the video test, the station classification model achieved a per-frame accuracy of 86.2%. Compared with EUS experts, the models achieved 90.0% accuracy in classification and .77 and .813 dice in blood vessel and pancreas segmentation, which is comparable with that of experts. In the crossover study, trainee station recognition accuracy improved from 67.2% to 78.4% (95% confidence interval, .058-1.663; P < .01). CONCLUSIONS: The BP MASTER system has the potential to play an important role in shortening the pancreatic EUS learning curve and improving EUS quality control in the future.

A novel artificial intelligence system for the assessment of bowel preparation (with video).

BACKGROUND AND AIMS: The quality of bowel preparation is an important factor that can affect the effectiveness of a colonoscopy. Several tools, such as the Boston Bowel Preparation Scale (BBPS) and Ottawa Bowel Preparation Scale, have been developed to evaluate bowel preparation. However, understanding the differences between evaluation methods and consistently applying them can be challenging for endoscopists. There are also subjective biases and differences among endoscopists. Therefore, this study aimed to develop a novel, objective, and stable method for the assessment of bowel preparation through artificial intelligence. METHODS: We used a deep convolutional neural network to develop this novel system. First, we retrospectively collected colonoscopy images to train the system and then compared its performance with endoscopists via a human-machine contest. Then, we applied this model to colonoscopy videos and developed a system named ENDOANGEL to provide bowel preparation scores every 30 seconds and to show the cumulative ratio of frames for each score during the withdrawal phase of the colonoscopy. RESULTS: ENDOANGEL achieved 93.33% accuracy in the human-machine contest with 120 images, which was better than that of all endoscopists. Moreover, ENDOANGEL achieved 80.00% accuracy among 100 images with bubbles. In 20 colonoscopy videos, accuracy was 89.04%, and ENDOANGEL continuously showed the accumulated percentage of the images for different BBPS scores during the withdrawal phase and prompted us for bowel preparation scores every 30 seconds. CONCLUSIONS: We provided a novel and more accurate evaluation method for bowel preparation and developed an objective and stable system-ENDOANGEL-that could be applied reliably and steadily in clinical settings.

A deep learning-based system for mediastinum station localization in linear EUS (with video).

Background and Objectives: EUS is a crucial diagnostic and therapeutic method for many anatomical regions, especially in the evaluation of mediastinal diseases and related pathologies. Rapidly finding the standard stations is the key to achieving efficient and complete mediastinal EUS imaging. However, it requires substantial technical skills and extensive knowledge of mediastinal anatomy. We constructed a system, named EUS-MPS (EUS-mediastinal position system), for real-time mediastinal EUS station recognition. Methods: The standard scanning of mediastinum EUS was divided into 7 stations. There were 33 010 images in mediastinum EUS examination collected to construct a station classification model. Then, we used 151 videos clips for video validation and used 1212 EUS images from 2 other hospitals for external validation. An independent data set containing 230 EUS images was applied for the man-machine contest. We conducted a crossover study to evaluate the effectiveness of this system in reducing the difficulty of mediastinal ultrasound image interpretation. Results: For station classification, the model achieved an accuracy of 90.49% in image validation and 83.80% in video validation. At external validation, the models achieved 89.85% accuracy. In the man-machine contest, the model achieved an accuracy of 84.78%, which was comparable to that of expert (83.91%). The accuracy of the trainees' station recognition was significantly improved in the crossover study, with an increase of 13.26% (95% confidence interval, 11.04%-15.48%; P < 0.05). Conclusions: This deep learning-based system shows great performance in mediastinum station localization, having the potential to play an important role in shortening the learning curve and establishing standard mediastinal scanning in the future.

A deep learning-based system to identify originating mural layer of upper gastrointestinal submucosal tumors under EUS.

Background and Objective: EUS is the most accurate procedure to determine the originating mural layer and subsequently select the treatment of submucosal tumors (SMTs). However, it requires superb technical and cognitive skills. In this study, we propose a system named SMT Master to determine the originating mural layer of SMTs under EUS. Materials and Methods: We developed 3 models: deep convolutional neural network (DCNN) 1 for lesion segmentation, DCNN2 for mural layer segmentation, and DCNN3 for the originating mural layer classification. A total of 2721 EUS images from 201 patients were used to train the 3 models. We validated our model internally and externally using 283 images from 26 patients and 172 images from 26 patients, respectively. We applied 368 images from 30 patients for the man-machine contest and used 30 video clips to test the originating mural layer classification. Results: In the originating mural layer classification task, DCNN3 achieved a classification accuracy of 84.43% and 80.68% at internal and external validations, respectively. In the video test, the accuracy was 80.00%. DCNN1 achieved Dice coefficients of 0.956 and 0.776 for lesion segmentation at internal and external validations, respectively, whereas DCNN2 achieved Dice coefficients of 0.820 and 0.740 at internal and external validations, respectively. The system achieved 90.00% accuracy in classification, which is comparable with that of EUS experts. Conclusions: Our proposed system has the potential to solve difficulties in determining the originating mural layer of SMTs in EUS procedures, which relieves the EUS learning pressure of physicians.

Complementary Effect of the Proportion of Overspeed Frames of Withdrawal and Withdrawal Time on Reflecting Colonoscopy Quality: A Retrospective, Observational Study.

INTRODUCTION: Constructing quality indicators that reflect the defect of colonoscopy operation for quality audit and feedback is very important. Previously, we have established a real-time withdrawal speed monitoring system to control withdrawal speed below the safe speed. We aimed to explore the relationship between the proportion of overspeed frames (POF) of withdrawal and the adenoma detection rate (ADR) and to conjointly analyze the influence of POF and withdrawal time on ADR to evaluate the feasibility of POF combined with withdrawal time as a quality control indicator. METHODS: The POF was defined as the proportion of frames with instantaneous speed ≥44 in the whole colonoscopy video. First, we developed a system for the POF of withdrawal based on a perceptual hashing algorithm. Next, we retrospectively collected 1,804 colonoscopy videos to explore the relationship between POF and ADR. According to withdrawal time and POF cutoff, we conducted a complementary analysis on the effects of POF and withdrawal time on ADR. RESULTS: There was an inverse correlation between the POF and ADR (Pearson correlation coefficient -0.836). When withdrawal time was >6 minutes, the ADR of the POF ≤10% was significantly higher than that of POF >10% (25.30% vs 16.50%; odds ratio 0.463, 95% confidence interval 0.296-0.724, P < 0.01). When the POF was ≤10%, the ADR of withdrawal time >6 minutes was higher than that of withdrawal time ≤6 minutes (25.30% vs 21.14%; odds ratio 0.877, 95% confidence interval 0.667-1.153, P = 0.35). DISCUSSION: The POF was strongly correlated with ADR. The combined assessment of the POF and withdrawal time has profound significance for colonoscopy quality control.

An explainable artificial intelligence system for diagnosing Helicobacter Pylori infection under endoscopy: a case-control study.

Background: Changes in gastric mucosa caused by Helicobacter pylori (H. pylori) infection affect the observation of early gastric cancer under endoscopy. Although previous researches reported that computer-aided diagnosis (CAD) systems have great potential in the diagnosis of H. pylori infection, their explainability remains a challenge. Objective: We aim to develop an explainable artificial intelligence system for diagnosing H. pylori infection (EADHI) and giving diagnostic basis under endoscopy. Design: A case-control study. Methods: We retrospectively obtained 47,239 images from 1826 patients between 1 June 2020 and 31 July 2021 at Renmin Hospital of Wuhan University for the development of EADHI. EADHI was developed based on feature extraction combining ResNet-50 and long short-term memory networks. Nine endoscopic features were used for H. pylori infection. EADHI's performance was evaluated and compared to that of endoscopists. An external test was conducted in Wenzhou Central Hospital to evaluate its robustness. A gradient-boosting decision tree model was used to examine the contributions of different mucosal features for diagnosing H. pylori infection. Results: The system extracted mucosal features for diagnosing H. pylori infection with an overall accuracy of 78.3% [95% confidence interval (CI): 76.2-80.3]. The accuracy of EADHI for diagnosing H. pylori infection (91.1%, 95% CI: 85.7-94.6) was significantly higher than that of endoscopists (by 15.5%, 95% CI: 9.7-21.3) in internal test. And it showed a good accuracy of 91.9% (95% CI: 85.6-95.7) in external test. Mucosal edema was the most important diagnostic feature for H. pylori positive, while regular arrangement of collecting venules was the most important H. pylori negative feature. Conclusion: The EADHI discerns H. pylori gastritis with high accuracy and good explainability, which may improve the trust and acceptability of endoscopists on CADs. Plain language summary: An explainable AI system for Helicobacter pylori with good diagnostic performance Helicobacter pylori (H. pylori) is the main risk factor for gastric cancer (GC), and changes in gastric mucosa caused by H. pylori infection affect the observation of early GC under endoscopy. Therefore, it is necessary to identify H. pylori infection under endoscopy. Although previous research showed that computer-aided diagnosis (CAD) systems have great potential in H. pylori infection diagnosis, their generalization and explainability are still a challenge. Herein, we constructed an explainable artificial intelligence system for diagnosing H. pylori infection (EADHI) using images by case. In this study, we integrated ResNet-50 and long short-term memory (LSTM) networks into the system. Among them, ResNet50 is used for feature extraction, LSTM is used to classify H. pylori infection status based on these features. Furthermore, we added the information of mucosal features in each case when training the system so that EADHI could identify and output which mucosal features are contained in a case. In our study, EADHI achieved good diagnostic performance with an accuracy of 91.1% [95% confidence interval (CI): 85.7-94.6], which was significantly higher than that of endoscopists (by 15.5%, 95% CI: 9.7-21.3%) in internal test. In addition, it showed a good diagnostic accuracy of 91.9% (95% CI: 85.6-95.7) in external tests. The EADHI discerns H. pylori gastritis with high accuracy and good explainability, which may improve the trust and acceptability of endoscopists on CADs. However, we only used data from a single center to develop EADHI, and it was not effective in identifying past H. pylori infection. Future, multicenter, prospective studies are needed to demonstrate the clinical applicability of CADs.

Assessment of the Role of Artificial Intelligence in the Association Between Time of Day and Colonoscopy Quality.

Importance: Time of day was associated with a decline in adenoma detection during colonoscopy. Artificial intelligence (AI) systems are effective in improving the adenoma detection rate (ADR), but the performance of AI during different times of the day remains unknown. Objective: To validate whether the assistance of an AI system could overcome the time-related decline in ADR during colonoscopy. Design, Setting, and Participants: This cohort study is a secondary analysis of 2 prospective randomized controlled trials (RCT) from Renmin Hospital of Wuhan University. Consecutive patients undergoing colonoscopy were randomly assigned to either the AI-assisted group or unassisted group from June 18, 2019, to September 6, 2019, and July 1, 2020, to October 15, 2020. The ADR of early and late colonoscopy sessions per half day were compared before and after the intervention of the AI system. Data were analyzed from March to June 2022. Exposure: Conventional colonoscopy or AI-assisted colonoscopy. Main Outcomes and Measures: Adenoma detection rate. Results: A total of 1780 patients (mean [SD] age, 48.61 [13.35] years, 837 [47.02%] women) were enrolled. A total of 1041 procedures (58.48%) were performed in early sessions, with 357 randomized into the unassisted group (34.29%) and 684 into the AI group (65.71%). A total of 739 procedures (41.52%) were performed in late sessions, with 263 randomized into the unassisted group (35.59%) and 476 into the AI group (64.41%). In the unassisted group, the ADR in early sessions was significantly higher compared with that of late sessions (13.73% vs 5.70%; P = .005; OR, 2.42; 95% CI, 1.31-4.47). After the intervention of the AI system, as expected, no statistically significant difference was found (22.95% vs 22.06%, P = .78; OR, 0.96; 95% CI; 0.71-1.29). Furthermore, the AI systems showed better assistance ability on ADR in late sessions compared with early sessions (odds ratio, 3.81; 95% CI, 2.10-6.91 vs 1.60; 95% CI, 1.10-2.34). Conclusions and Relevance: In this cohort study, AI systems showed higher assistance ability in late sessions per half day, which suggests the potential to maintain high quality and homogeneity of colonoscopies and further improve endoscopist performance in large screening programs and centers with high workloads.