Performance Comparison of the Deep Learning and the Human Endoscopist for Bleeding Peptic Ulcer Disease

Artificial intelligence-assisted system for the assessment of Forrest classification of peptic ulcer bleeding: a multicenter diagnostic study

BACKGROUND

Inaccurate Forrest classification may significantly affect clinical outcomes, especially in high risk patients. Therefore, this study aimed to develop a real-time deep convolutional neural network (DCNN) system to assess the Forrest classification of peptic ulcer bleeding (PUB).

METHODS

A training dataset (3868 endoscopic images) and an internal validation dataset (834 images) were retrospectively collected from the 900th Hospital, Fuzhou, China. In addition, 521 images collected from four other hospitals were used for external validation. Finally, 46 endoscopic videos were prospectively collected to assess the real-time diagnostic performance of the DCNN system, whose diagnostic performance was also prospectively compared with that of three senior and three junior endoscopists.

RESULTS

The DCNN system had a satisfactory diagnostic performance in the assessment of Forrest classification, with an accuracy of 91.2% (95%CI 89.5%-92.6%) and a macro-average area under the receiver operating characteristic curve of 0.80 in the validation dataset. Moreover, the DCNN system could judge suspicious regions automatically using Forrest classification in real-time videos, with an accuracy of 92.0% (95%CI 80.8%-97.8%). The DCNN system showed more accurate and stable diagnostic performance than endoscopists in the prospective clinical comparison test. This system helped to slightly improve the diagnostic performance of senior endoscopists and considerably enhance that of junior endoscopists.

CONCLUSION

The DCNN system for the assessment of the Forrest classification of PUB showed satisfactory diagnostic performance, which was slightly superior to that of senior endoscopists. It could therefore effectively assist junior endoscopists in making such diagnoses during gastroscopy.

Research trends on artificial intelligence and endoscopy in digestive diseases: A bibliometric analysis from 1990 to 2022

BACKGROUND

Recently, artificial intelligence (AI) has been widely used in gastrointestinal endoscopy examinations.

AIM

To comprehensively evaluate the application of AI-assisted endoscopy in detecting different digestive diseases using bibliometric analysis.

METHODS

Relevant publications from the Web of Science published from 1990 to 2022 were extracted using a combination of the search terms “AI” and “endoscopy”. The following information was recorded from the included publications: Title, author, institution, country, endoscopy type, disease type, performance of AI, publication, citation, journal and H-index.

RESULTS

A total of 446 studies were included. The number of articles reached its peak in 2021, and the annual citation numbers increased after 2006. China, the United States and Japan were dominant countries in this field, accounting for 28.7%, 16.8%, and 15.7% of publications, respectively. The Tada Tomohiro Institute of Gastroenterology and Proctology was the most influential institution. “Cancer” and “polyps” were the hotspots in this field. Colorectal polyps were the most concerning and researched disease, followed by gastric cancer and gastrointestinal bleeding. Conventional endoscopy was the most common type of examination. The accuracy of AI in detecting Barrett’s esophagus, colorectal polyps and gastric cancer from 2018 to 2022 is 87.6%, 93.7% and 88.3%, respectively. The detection rates of adenoma and gastrointestinal bleeding from 2018 to 2022 are 31.3% and 96.2%, respectively.

CONCLUSION

AI could improve the detection rate of digestive tract diseases and a convolutional neural network-based diagnosis program for endoscopic images shows promising results.

An Improved Endoscopic Automatic Classification Model for Gastroesophageal Reflux Disease Using Deep Learning Integrated Machine Learning

Gastroesophageal reflux disease (GERD) is a common digestive tract disease, and most physicians use the Los Angeles classification and diagnose the severity of the disease to provide appropriate treatment. With the advancement of artificial intelligence, deep learning models have been used successfully to help physicians with clinical diagnosis. This study combines deep learning and machine learning techniques and proposes a two-stage process for endoscopic classification in GERD, including transfer learning techniques applied to the target dataset to extract more precise image features and machine learning algorithms to build the best classification model. The experimental results demonstrate that the performance of the GerdNet-RF model proposed in this work is better than that of previous studies. Test accuracy can be improved from 78.8% ± 8.5% to 92.5% ± 2.1%. By enhancing the automated diagnostic capabilities of AI models, patient health care will be more assured.

Chemoprevention of Colitis-Associated Dysplasia or Cancer in Inflammatory Bowel Disease

The association between inflammatory bowel disease and colorectal cancer is well known. Although the overall incidence of inflammatory bowel disease has declined recently, patients with this disease still have a 1.7-fold increased risk of colorectal cancer. The risk factors for developing colorectal cancer include extensive colitis, young age at diagnosis, disease duration, primary sclerosing cholangitis, chronic colonic mucosal inflammation, dysplasia lesion, and post-inflammatory polyps. In patients with inflammatory bowel disease, control of chronic inflammation and surveillance colonoscopies are important for the prevention of colorectal cancer. The 2017 guidelines from the European Crohn's and Colitis Organisation suggest that colonoscopies to screen for colorectal cancer should be performed when inflammatory bowel disease symptoms have lasted for 8 years. Current evidence supports the use of chemoprevention therapy with mesalamine to reduce the risk of colorectal cancer in patients with ulcerative colitis. Other compounds, including thiopurine, folic acid, statin, and tumor necrosis factor-α inhibitor, are controversial. Large surveillance cohort studies with longer follow-up duration are needed to evaluate the impact of drugs on colorectal cancer risks.

Development and validation of a deep learning-based algorithm for colonoscopy quality assessment

BACKGROUND

Quality indicators should be assessed and monitored to improve colonoscopy quality in clinical practice. Endoscopists must enter relevant information in the endoscopy reporting system to facilitate data collection, which may be inaccurate. The current study aimed to develop a full deep learning-based algorithm to identify and analyze intra-procedural colonoscopy quality indicators based on endoscopy images obtained during the procedure.

METHODS

A deep learning system for classifying colonoscopy images for quality assurance purposes was developed and its performance was assessed with an independent dataset. The system was utilized to analyze captured images and results were compared with those of real-world reports.

RESULTS

In total, 10,417 images from the hospital endoscopy database and 3157 from Hyper-Kvasir open dataset were utilized to develop the quality assurance algorithm. The overall accuracy of the algorithm was 96.72% and that of the independent test dataset was 94.71%. Moreover, 761 real-world reports and colonoscopy images were analyzed. The accuracy of electronic reports about cecal intubation rate was 99.34% and that of the algorithm was 98.95%. The agreement rate for the assessment of polypectomy rates using the electronic reports and the algorithm was 0.87 (95% confidence interval 0.83-0.90). A good correlation was found between the withdrawal time calculated using the algorithm and that entered by the physician (correlation coefficient r = 0.959, p < 0.0001).

CONCLUSION

We proposed a novel deep learning-based algorithm that used colonoscopy images for quality assurance purposes. This model can be used to automatically assess intra-procedural colonoscopy quality indicators in clinical practice.

Comparison of Adverse Events of Different Endoscopic Ultrasound-Guided Tissue Acquisition Methods: A Single-Center Retrospective Analysis

The efficacy of new generation endoscopic ultrasound-guided biopsy needles has been promising in recent years. Yet, comparing these needles’ diagnostic yield and safety to conventional needles is not well-known. Our study aims to compare the adverse events of endoscopic ultrasound-guided tissue acquisition (EUS-TA) with different types of needles, including FNA needles, FNB needles with a Franseen tip and FNB needles with a reverse bevel. Furthermore, we will analyze the risk factors, including tumor vascularity, different needle types, and the underlying disease, which may impact the safety of the procedures. From May 2014 to December 2021, 192 consecutive EUS-TAs were performed on pancreatic and peripancreatic lesions in our hospital using different types of FNA and FNB needles. We retrospectively reviewed the data and identified the risk factors for EUS-TA-related complications. As a result, the hypervascular tumor is a significant risk factor for adverse events in our multivariate analysis, with an odds ratio of 4.96 (95% CI 1.33–18.47), while liver cirrhosis is one of the risk factors for adverse events during EUS-TA, with an odds ratio of 5.3 (95% CI 1.1–25.6). However, the risk of adverse events did not increase using Franseen-tip needles, compared to conventional FNA or FNB needles with a reverse bevel. In conclusion, we must be more cautious in patients with liver cirrhosis and hypervascular tumors, such as pancreatic neuroendocrine tumors, when performing EUS-guided tissue acquisition.

Comparison Between Same-Day and Split-Dose Preparations with Sodium Picosulfate/Magnesium Citrate: A Randomized Noninferiority Study

BACKGROUND

Sodium picosulfate/magnesium citrate (SPMC) is a small-volume bowel cleansing agent with similar efficacy to and better tolerability than polyethylene glycol. However, we found no data on which SPMC preparation (same-day vs. split-dose) provides better bowel cleansing efficacy for afternoon colonoscopy.

AIMS

To compare bowel cleansing efficacy of different timing of the regimen.

METHODS

This randomized, single-center, endoscopist-blinded, noninferior study compared same-day and split-dose SPMC preparations for afternoon colonoscopy in 101 and 96 patients, respectively. We also included a prospective observation group of 100 patients receiving morning colonoscopy to compare bowel preparation between morning and afternoon colonoscopies. Bowel cleansing efficacy was then evaluated by the Aronchick Scale, Ottawa Bowel Preparation Scale (OBPS), Boston Bowel Preparation Scale (BBPS), and the Bubble Scale.

RESULTS

Same-day and split-dose preparations were similar in efficacy in all four scales. In the Aronchick Scale, the success rate (excellent and good cleanliness) was higher in same-day preparation than in split-dose preparation (100% vs. 92.8%). The same-day preparation also obtained a better OBPS score (1.4 vs. 2.1), but BBPS showed no difference between such groups (7.7 vs. 7.4).

CONCLUSION

Same-day preparation with SPMC is not inferior to split-dose preparation for afternoon colonoscopy.

Upper endoscopy photodocumentation quality evaluation with novel deep learning system

OBJECTIVES

Visualization and photodocumentation during endoscopy procedures are suggested to be one indicator for endoscopy performance quality. However, this indicator is difficult to measure and audit manually in clinical practice. Artificial intelligence (AI) is an emerging technology that may solve this problem.

METHODS

A deep learning model with an accuracy of 96.64% was developed from 15,305 images for upper endoscopy anatomy classification in the unit. Endoscopy images for asymptomatic patients receiving screening endoscopy were evaluated with this model to assess the completeness of photodocumentation rate.

RESULTS

A total of 15,723 images from 472 upper endoscopies performed by 12 endoscopists were enrolled. The complete photodocumentation rate from the pharynx to the duodenum was 53.8% and from the esophagus to the duodenum was 78.0% in this study. Endoscopists with a higher adenoma detection rate had a higher complete examination rate from the pharynx to duodenum (60.0% vs. 38.7%, P < 0.0001) and from esophagus to duodenum (83.0% vs. 65.7%, P < 0.0001) compared with endoscopists with lower adenoma detection rate. The pharynx, gastric angle, gastric retroflex view, gastric antrum, and the first portion of duodenum are likely to be missed by endoscopists with lower adenoma detection rates.

CONCLUSIONS

We report the use of a deep learning model to audit endoscopy photodocumentation quality in our unit. Endoscopists with better performance in colonoscopy had a better performance for this quality indicator. The use of such an AI system may help the endoscopy unit audit endoscopy performance.

Machine-Learning Algorithm for Predicting Fatty Liver Disease in a Taiwanese Population

The rising incidence of fatty liver disease (FLD) poses a health challenge, and is expected to be the leading global cause of liver-related morbidity and mortality in the near future. Early case identification is crucial for disease intervention. A retrospective cross-sectional study was performed on 31,930 Taiwanese subjects (25,544 training and 6386 testing sets) who had received health check-ups and abdominal ultrasounds in Changhua Christian Hospital from January 2009 to January 2019. Clinical and laboratory factors were included for analysis by different machine-learning algorithms. In addition, the performance of the machine-learning algorithms was compared with that of the fatty liver index (FLI). Totally, 6658/25,544 (26.1%) and 1647/6386 (25.8%) subjects had moderate-to-severe liver disease in the training and testing sets, respectively. Five machine-learning models were examined and demonstrated exemplary performance in predicting FLD. Among these models, the xgBoost model revealed the highest area under the receiver operating characteristic (AUROC) (0.882), accuracy (0.833), F1 score (0.829), sensitivity (0.833), and specificity (0.683) compared with those of neural network, logistic regression, random forest, and support vector machine-learning models. The xgBoost, neural network, and logistic regression models had a significantly higher AUROC than that of FLI. Body mass index was the most important feature to predict FLD according to the feature ranking scores. The xgBoost model had the best overall prediction ability for diagnosing FLD in our study. Machine-learning algorithms provide considerable benefits for screening candidates with FLD.

Deep learning-based endoscopic anatomy classification: an accelerated approach for data preparation and model validation

BACKGROUND

Photodocumentation during endoscopy procedures is one of the indicators for endoscopy performance quality; however, this indicator is difficult to measure and audit in the endoscopy unit. Emerging artificial intelligence technology may solve this problem, which requires a large amount of material for model development. We developed a deep learning-based endoscopic anatomy classification system through convolutional neural networks with an accelerated data preparation approach.

PATIENTS AND METHODS

We retrospectively collected 8,041 images from esophagogastroduodenoscopy (EGD) procedures and labeled them using two experts for nine anatomical locations of the upper gastrointestinal tract. A base model for EGD image multiclass classification was first developed, and an additional 6,091 images were enrolled and classified by the base model. A total of 5,963 images were manually confirmed and added to develop the subsequent enhanced model. Additional internal and external endoscopy image datasets were used to test the model performance.

RESULTS

The base model achieved total accuracy of 96.29%. For the enhanced model, the total accuracy was 96.64%. The overall accuracy improved with the enhanced model compared with the base model for the internal test dataset without narrowband images (93.05% vs. 91.25%, p < 0.01) or with narrowband images (92.74% vs. 90.46%, p < 0.01). The total accuracy was 92.56% of the enhanced model on the external test dataset.

CONCLUSIONS

We constructed a deep learning-based model with an accelerated approach that can be used for quality control in endoscopy units. The model was also validated with both internal and external datasets with high accuracy.

Forrest Classification for Bleeding Peptic Ulcer: A New Look at the Old Endoscopic Classification

The management of peptic ulcer bleeding is clinically challenging. For decades, the Forrest classification has been used for risk stratification for nonvariceal ulcer bleeding. The perception and interpretation of the Forrest classification vary among different endoscopists. The relationship between the bleeder and ulcer images and the different stages of the Forrest classification has not been studied yet. Endoscopic still images of 276 patients with peptic ulcer bleeding for the past 3 years were retrieved and reviewed. The intra-rater agreement and inter-rater agreement were compared. The obtained endoscopic images were manually drawn to delineate the extent of the ulcer and bleeding area. The areas of the region of interest were compared between the different stages of the Forrest classification. A total of 276 images were first classified by two experienced tutor endoscopists. The images were reviewed by six other endoscopists. A good intra-rater correlation was observed (0.92–0.98). A good inter-rater correlation was observed among the different levels of experience (0.639–0.859). The correlation was higher among tutor and junior endoscopists than among experienced endoscopists. Low-risk Forrest IIC and III lesions show distinct patterns compared to high-risk Forrest I, IIA, or IIB lesions. We found good agreement of the Forrest classification among different endoscopists in a single institution. This is the first study to quantitively analyze the obtained and explain the distinct patterns of bleeding ulcers from endoscopy images.

Application Status and Prospects of Artificial Intelligence in Peptic Ulcers

Peptic ulcer (PU) is a common and frequently occurring disease. Although PU seriously threatens the lives and health of global residents, the applications of artificial intelligence (AI) have strongly promoted diversification and modernization in the diagnosis and treatment of PU. This minireview elaborates on the research progress of AI in the field of PU, from PU's pathogenic factor Helicobacter pylori (Hp) infection, diagnosis and differential diagnosis, to its management and complications (bleeding, obstruction, perforation and canceration). Finally, the challenges and prospects of AI application in PU are prospected and expounded. With the in-depth understanding of modern medical technology, AI remains a promising option in the management of PU patients and plays a more indispensable role. How to realize the robustness, versatility and diversity of multifunctional AI systems in PU and conduct multicenter prospective clinical research as soon as possible are the top priorities in the future.

A Nomogram for Predicting Laparoscopic and Endoscopic Cooperative Surgery during the Endoscopic Resection of Subepithelial Tumors of the Upper Gastrointestinal Tract

Background: Considering the widespread use of esophagogastroduodenoscopy, the prevalence of upper gastrointestinal (GI) subepithelial tumors (SET) increases. For relatively safer removal of upper GI SETs, endoscopic submucosal dissection (ESD) has been developed as an alternative to surgery. This study aimed to analyze the outcome of endoscopic resection for SETs and develop a prediction model for the need for laparoscopic and endoscopic cooperative surgery (LECS) during the procedure. Method: We retrospectively analyzed 123 patients who underwent endoscopic resection for upper GI SETs between January 2012 and December 2020 at our institution. Intraoperatively, they underwent ESD or submucosal tunneling endoscopic resection (STER). Results: ESD and STER were performed in 107 and 16 patients, respectively. The median age was 55 years, and the average tumor size was 1.5 cm. En bloc resection was achieved in 114 patients (92.7%). The median follow-up duration was 242 days without recurrence. Perforation occurred in 47 patients (38.2%), and 30 patients (24.4%) underwent LECS. Most perforations occurred in the fundus. Through multivariable analysis, we built a nomogram that can predict LECS requirement according to tumor location, size, patient age, and sex. The prediction model exhibited good discrimination ability, with an area under the curve (AUC) of 0.893. Conclusions: Endoscopic resection is a noninvasive procedure for small upper-GI SETs. Most perforations can be successfully managed endoscopically. The prediction model for LECS requirement is useful in treatment planning.

Real-World Experience of Chronic Hepatitis C-Related Compensated Liver Cirrhosis Treated with Glecaprevir/Pibrentasvir: A Multicenter Retrospective Study

Background: Glecaprevir/pibrentasvir is a protease inhibitor-containing pangenotypic direct-acting antiviral regimen that has been approved for the treatment of chronic hepatitis C. The present study aimed to evaluate the safety and efficacy of glecaprevir/pibrentasvir in patients with compensated cirrhosis in a real-world setting. Methods: We evaluated the real-world safety and efficacy of glecaprevir/pibrentasvir in patients with compensated cirrhosis from five hospitals in the Changhua Christian Care System, who underwent treatment between August 2018 and October 2020. The primary endpoint was a sustained virological response observed 12 weeks after completion of the treatment. Results: Ninety patients, including 70 patients who received the 12-week therapy and 20 patients who received the 8-week therapy, were enrolled. The mean age of the patients was 65 years, and 57.8% of the patients were males. Sixteen (17.8%) patients had end-stage renal disease, and 15 (16.7%) had co-existing hepatoma. The hepatitis C virus genotypes 1 (40%) and 2 (35.6%) were most common. The common side effects included anorexia (12.2%), pruritus (7.8%), abdominal discomfort (7.8%), and malaise (7.8%). Laboratory adverse grade ≥3 events included anemia (6.3%), thrombocytopenia (5.1%), and jaundice (2.2%). The overall sustained virological response rates were 94.4% and 97.7% in the intention-to-treat and per-protocol analyses, respectively. Conclusions: the glecaprevir/pibrentasvir treatment regimen was highly effective and well tolerated among patients with compensated cirrhosis in the real-world setting.

Comparison of the sixth, seventh, and eighth editions of the American Joint Committee on Cancer Tumor-Node-Metastasis staging system for gastric cancer: A single institution experience

In 2018, the eighth edition of the American Joint Committee on Cancer Tumor-Node-Metastasis classification and staging system was implemented. Few reports were made comparing the performance of different editions of the American Joint Committee on Cancer (AJCC) system. Therefore, this study aimed to examine the prognostic predictability from the sixth to the eighth editions of the AJCC staging system for gastric cancer.A total of 414 patients with gastric cancer who underwent surgery at Changhua Christian Hospital from January 2007 to December 2017 were enrolled in the study. To identify the prognostic factors for gastric cancer death, univariate and multivariate analyses were performed. The homogeneity and discrimination abilities of the sixth to eighth editions of the staging system were compared using the likelihood ratio chi-square test, linear trend chi-square test, and Akaike information criterion.The sixth edition of the staging system had the lowest Akaike information criterion value, suggesting a better prognostic stratification than other editions. From the result of the likelihood ratio chi-square test, the T and N staging systems of the seventh and eighth editions had better homogeneity and discriminatory ability than the sixth edition. The eighth edition had better prognostic performance in patients at stage III compared with the seventh edition.The AJCC seventh and eighth editions had improved prognostic predictability of the T and N factors compared with the sixth edition. However, the overall staging performance of the eighth edition is not superior compared to the sixth edition. Further studies with larger sample size should be conducted to compare the performance of different editions of the AJCC staging system for different ethnic populations.

Endoscopic Resection for Gastric Subepithelial Tumor with Backup Laparoscopic Surgery: Description of a Single-Center Experience

The aim of this study was to analyze patients who underwent endoscopic resection (ER) for gastric subepithelial tumors (SETs) with a high probability of surgical intervention. Between January 2013 and January 2021, 83 patients underwent ER at the operation theater and 27 patients (32.5%) required backup surgery mainly due to incidental perforation or uncontrolled bleeding despite endoscopic repairing. The tumor was predominantly located in the upper-third stomach (81%) with a size ≤ 2 cm (69.9%) and deep to the muscularis propria (MP) layer (92.8%) but there were no significant differences between two groups except tumor exophytic growth as a risk factor in the surgery group (37% vs. 0%, p < 0.0001). Patients in the ER-only group had shorter durations of procedure times (60 min vs. 185 min, p < 0.0001) and lengths of stay (5 days vs. 7 days, p < 0.0001) but with a higher percentage of overall morbidity graded III (0% vs. 7.1%, p = 0.1571). After ER, five patients (6%) had delayed perforation and two (2.4%) required emergent laparoscopic surgery. Neither recurrence nor gastric stenosis was reported during long-term surveillance. Here, we provide a minimally invasive strategy of endoscopic resection with backup laparoscopic surgery for gastric SETs.

Current Status and Future Perspective of Artificial Intelligence in the Management of Peptic Ulcer Bleeding: A Review of Recent Literature

With the decreasing incidence of peptic ulcer bleeding (PUB) over the past two decades, the clinician experience of managing patients with PUB has also declined, especially for young endoscopists. A patient with PUB management requires collaborative care involving the emergency department, gastroenterologist, radiologist, and surgeon, from initial assessment to hospital discharge. The application of artificial intelligence (AI) methods has remarkably improved people's lives. In particular, AI systems have shown great potential in many areas of gastroenterology to increase human performance. Colonoscopy polyp detection or diagnosis by an AI system was recently introduced for commercial use to improve endoscopist performance. Although PUB is a longstanding health problem, these newly introduced AI technologies may soon impact endoscopists' clinical practice by improving the quality of care for these patients. To update the current status of AI application in PUB, we reviewed recent relevant literature and provided future perspectives that are required to integrate such AI tools into real-world practice.