Artificial intelligence in endoscopy

Can artificial intelligence replace endoscopists when assessing mucosal healing in ulcerative colitis? A systematic review and diagnostic test accuracy meta-analysis

BACKGROUNDS AND AIMS

Mucosal healing (MH) in inflammatory bowel diseases (IBD) is an important landmark for clinical decision making. Artificial intelligence systems (AI) that automatically deliver the grade of endoscopic inflammation may solve moderate interobserver agreement and the need of central reading in clinical trials.

METHODS

We performed a systematic review of EMBASE and MEDLINE databases up to 01/12/2022 following PRISMA and the Joanna Briggs Institute methodologies to answer the following question: "Can AI replace endoscopists when assessing MH in IBD?". The research was restricted to ulcerative colitis (UC), and a diagnostic odds ratio (DOR) meta-analysis was performed. Risk of bias was evaluated with QUADAS-2 tool.

RESULTS

A total of 21 / 739 records were selected for full text evaluation, and 12 were included in the meta-analysis. Deep learning algorithms based on convolutional neural networks architecture achieved a satisfactory performance in evaluating MH on UC, with sensitivity, specificity, DOR and SROC of respectively 0.91(CI95 %:0.86-0.95);0.89(CI95 %:0.84-0.93);92.42(CI95 %:54.22-157.53) and 0.957 when evaluating fixed images (n = 8) and 0.86(CI95 %:0.75-0.93);0.91(CI95 %:0.87-0.94);70.86(CI95 %:24.63-203.86) and 0.941 when evaluating videos (n = 6). Moderate-high levels of heterogeneity were noted, limiting the quality of the evidence.

CONCLUSIONS

AI systems showed high potential in detecting MH in UC with optimal diagnostic performance, although moderate-high heterogeneity of the data was noted. Standardised and shared AI training may reduce heterogeneity between systems.

Application value of an artificial intelligence-based diagnosis and recognition system in gastroscopy training for graduate students in gastroenterology: a preliminary study

OBJECTIVE

This study aimed to discuss the application value of an artificial intelligence-based diagnosis and recognition system (AIDRS) in the teaching activities for Bachelor of Medicine and Bachelor of Surgery (MBBS) in China. The learning performance of graduate students in gastroenterology during gastroscopy training with and without AIDRS was assessed.

METHODS

The study recruited 32 graduate students of the gastroenterology program at Jiangsu province hospital of Chinese medicine and Xiangyang No. 1 People's Hospital from March 2018 to March 2022 and randomly divided them into AIDRS (n = 16) and non-AIDRS (n = 16) groups. The AIDRS software was used for real-time monitoring of blind spots of gastroscopy to aid in lesion diagnosis and recognition in the AIDRS group. Only a conventional gastroscopic procedure was implemented in the non-AIDRS group. The final performance score, success rate of gastroscopy, lesion detection rate, and pain score of patients were compared between the two groups during gastroscopy. A self-prepared teaching and learning satisfaction questionnaire was administered to the two groups of students.

RESULTS

The AIDRS group had a higher final performance score (92.60 ± 2.83 vs. 89.21 ± 3.57, t = 2.98, P < 0.05), a higher success rate of gastroscopy (448/480 vs. 417/480, χ2 = 11.23, P < 0.05), and a higher detection rate of lesions (51/52 vs. 41/53, χ2 = 8.56, P < 0.05) compared with the non-AIDRS group. The pain scores of patients were lower in the AIDRS group than in the non-AIDRS group (3.40 [2.23, 3.98] vs. 4.45 [3.72, 4.75], Z = 3.04, P < 0.05). Besides, the average time for gastroscopy was lower in the AIDRS group than in the non-AIDRS group (7.15 ± 1.24 vs. 8.21 ± 1.26, t = 2.38, P = 0.02). The overall satisfaction level with the teaching program was higher in the AIDRS group (43.51 ± 2.29 vs. 40.93 ± 2.07, t = 3.33, P < 0.05).

CONCLUSION

In the context of medicine-education cooperation, AIDRS offered valuable assistance in gastroscopy training and increased the success rate of gastroscopy and teaching and learning satisfaction. AIDRS is worthy of wider-scale promotion.

Development and evaluation of a program based on a generative pre-trained transformer model from a public natural language processing platform for efficiency enhancement in post-procedural quality control of esophageal endoscopic submucosal dissection

BACKGROUND

Post-procedural quality control of endoscopic submucosal dissection (ESD) is emphasized in guidelines. However, this process can be tedious and time-consuming. Recently, a pre-training model called generative pre-trained transformer (GPT) on a public natural language processing platform has emerged and garnered significant attention, whose capabilities align well with the post-procedural quality control process and have the potential to streamline it. Therefore, we developed a simple program utilizing this platform and evaluated its performance.

METHODS

Esophageal ESDs were retrospectively included. The manual quality control process was performed and act as reference standard. GPT's prompt was optimized through multiple iterations. A Python program was developed to automatically submit prompt with pathological report of each ESD procedure and collect quality control information provided by GPT. Its performance on quality control was evaluated with accuracy, precision, recall, and F-1 score.

RESULTS

165 cases were involved into the dataset, of which 5 were utilized as the prompt optimization dataset and 160 as the validation dataset. Definitive prompt was achieved through seven iterations. Time spent on the validation dataset by GPT was 13.47 ± 2.43 min. Accuracies of pathological diagnosis, invasion depth, horizontal margin, vertical margin, vascular invasion, and lymphatic invasion of the quality control program were (0.940, 0.952) (95% CI), (0.925, 0.945) (95% CI), 0.931, 1.0, and 1.0, respectively. Precisions were (0.965, 0.969) (95% CI), (0.934, 0.954) (95% CI), and 0.957 for pathological diagnosis, invasion depth, and horizontal margin, respectively. Recalls were (0.940, 0.952) (95% CI), (0.925, 0.945) (95% CI), and 0.931 for factors as mentioned, respectively. F1-score were (0.945, 0.957) (95% CI), (0.928, 0.948) (95% CI), and 0.941 for factors as mentioned, respectively.

CONCLUSIONS

This quality control program was qualified of post-procedural quality control of esophageal ESDs. GPT can be easily applied to this quality control process and reduce workload of the endoscopists.

Performance and comparison of artificial intelligence and human experts in the detection and classification of colonic polyps

OBJECTIVE

The main aim of this study was to analyze the performance of different artificial intelligence (AI) models in endoscopic colonic polyp detection and classification and compare them with doctors with different experience.

METHODS

We searched the studies on Colonoscopy, Colonic Polyps, Artificial Intelligence, Machine Learning, and Deep Learning published before May 2020 in PubMed, EMBASE, Cochrane, and the citation index of the conference proceedings. The quality of studies was assessed using the QUADAS-2 table of diagnostic test quality evaluation criteria. The random-effects model was calculated using Meta-DISC 1.4 and RevMan 5.3.

RESULTS

A total of 16 studies were included for meta-analysis. Only one study (1/16) presented externally validated results. The area under the curve (AUC) of AI group, expert group and non-expert group for detection and classification of colonic polyps were 0.940, 0.918, and 0.871, respectively. AI group had slightly lower pooled specificity than the expert group (79% vs. 86%, P < 0.05), but the pooled sensitivity was higher than the expert group (88% vs. 80%, P < 0.05). While the non-experts had less pooled specificity in polyp recognition than the experts (81% vs. 86%, P < 0.05), and higher pooled sensitivity than the experts (85% vs. 80%, P < 0.05).

CONCLUSION

The performance of AI in polyp detection and classification is similar to that of human experts, with high sensitivity and moderate specificity. Different tasks may have an impact on the performance of deep learning models and human experts, especially in terms of sensitivity and specificity.

Artificial Intelligence for Colonoscopy: Past, Present, and Future

During the past decades, many automated image analysis methods have been developed for colonoscopy. Real-time implementation of the most promising methods during colonoscopy has been tested in clinical trials, including several recent multi-center studies. All trials have shown results that may contribute to prevention of colorectal cancer. We summarize the past and present development of colonoscopy video analysis methods, focusing on two categories of artificial intelligence (AI) technologies used in clinical trials. These are (1) analysis and feedback for improving colonoscopy quality and (2) detection of abnormalities. Our survey includes methods that use traditional machine learning algorithms on carefully designed hand-crafted features as well as recent deep-learning methods. Lastly, we present the gap between current state-of-the-art technology and desirable clinical features and conclude with future directions of endoscopic AI technology development that will bridge the current gap.

Deep learning model for diagnosing gastric mucosal lesions using endoscopic images: development, validation, and method comparison

BACKGROUND AND AIMS

Endoscopic differential diagnoses of gastric mucosal lesions (benign gastric ulcer, early gastric cancer [EGC], and advanced gastric cancer) remain challenging. We aimed to develop and validate convolutional neural network-based artificial intelligence (AI) models: lesion detection, differential diagnosis (AI-DDx), and invasion depth (AI-ID; pT1a vs pT1b among EGC) models.

METHODS

This study included 1366 consecutive patients with gastric mucosal lesions from 2 referral centers in Korea. One representative endoscopic image from each patient was used. Histologic diagnoses were set as the criterion standard. Performance of the AI-DDx (training/internal/external validation set, 1009/112/245) and AI-ID (training/internal/external validation set, 620/68/155) was compared with visual diagnoses by independent endoscopists (stratified by novice [<1 year of experience], intermediate [2-3 years of experience], and expert [>5 years of experience]) and EUS results, respectively.

RESULTS

The AI-DDx showed good diagnostic performance for both internal (area under the receiver operating characteristic curve [AUROC] = .86) and external validation (AUROC = .86). The performance of the AI-DDx was better than that of novice (AUROC = .82, P = .01) and intermediate endoscopists (AUROC = .84, P = .02) but was comparable with experts (AUROC = .89, P = .12) in the external validation set. The AI-ID showed a fair performance in both internal (AUROC = .78) and external validation sets (AUROC = .73), which were significantly better than EUS results performed by experts (internal validation, AUROC = .62; external validation, AUROC = .56; both P < .001).

CONCLUSIONS

The AI-DDx was comparable with experts and outperformed novice and intermediate endoscopists for the differential diagnosis of gastric mucosal lesions. The AI-ID performed better than EUS for evaluation of invasion depth.

Survey on the perceptions of UK gastroenterologists and endoscopists to artificial intelligence

BACKGROUND AND AIMS

With the potential integration of artificial intelligence (AI) into clinical practice, it is essential to understand end users' perception of this novel technology. The aim of this study, which was endorsed by the British Society of Gastroenterology (BSG), was to evaluate the UK gastroenterology and endoscopy communities' views on AI.

METHODS

An online survey was developed and disseminated to gastroenterologists and endoscopists across the UK.

RESULTS

One hundred four participants completed the survey. Quality improvement in endoscopy (97%) and better endoscopic diagnosis (92%) were perceived as the most beneficial applications of AI to clinical practice. The most significant challenges were accountability for incorrect diagnoses (85%) and potential bias of algorithms (82%). A lack of guidelines (92%) was identified as the greatest barrier to adopting AI in routine clinical practice. Participants identified real-time endoscopic image diagnosis (95%) as a research priority for AI, while the most perceived significant barriers to AI research were funding (82%) and the availability of annotated data (76%). Participants consider the priorities for the BSG AI Task Force to be identifying research priorities (96%), guidelines for adopting AI devices in clinical practice (93%) and supporting the delivery of multicentre clinical trials (91%).

CONCLUSION

This survey has identified views from the UK gastroenterology and endoscopy community regarding AI in clinical practice and research, and identified priorities for the newly formed BSG AI Task Force.

'Artificial intelligence in Barrett's Esophagus'

Despite advances in endoscopic imaging modalities, there are still significant miss rates of dysplasia and cancer in Barrett's esophagus. Artificial intelligence (AI) is a promising tool that may potentially be a useful adjunct to the endoscopist in detecting subtle dysplasia and cancer. Studies have shown AI systems have a sensitivity of more than 90% and specificity of more than 80% in detecting Barrett's related dysplasia and cancer. Beyond visual detection and diagnosis, AI may also prove to be useful in quality control, streamlining clinical work, documentation, and lessening the administrative load on physicians. Research in this area is advancing at a rapid rate, and as the field expands, regulations and guidelines will need to be put into place to better regulate the growth and use of AI. This review provides an overview of the present and future role of AI in Barrett's esophagus.

Automatic Recognition of Colon and Esophagogastric Cancer with Machine Learning and Hyperspectral Imaging

There are approximately 1.8 million diagnoses of colorectal cancer, 1 million diagnoses of stomach cancer, and 0.6 million diagnoses of esophageal cancer each year globally. An automatic computer-assisted diagnostic (CAD) tool to rapidly detect colorectal and esophagogastric cancer tissue in optical images would be hugely valuable to a surgeon during an intervention. Based on a colon dataset with 12 patients and an esophagogastric dataset of 10 patients, several state-of-the-art machine learning methods have been trained to detect cancer tissue using hyperspectral imaging (HSI), including Support Vector Machines (SVM) with radial basis function kernels, Multi-Layer Perceptrons (MLP) and 3D Convolutional Neural Networks (3DCNN). A leave-one-patient-out cross-validation (LOPOCV) with and without combining these sets was performed. The ROC-AUC score of the 3DCNN was slightly higher than the MLP and SVM with a difference of 0.04 AUC. The best performance was achieved with the 3DCNN for colon cancer and esophagogastric cancer detection with a high ROC-AUC of 0.93. The 3DCNN also achieved the best DICE scores of 0.49 and 0.41 on the colon and esophagogastric datasets, respectively. These scores were significantly improved using a patient-specific decision threshold to 0.58 and 0.51, respectively. This indicates that, in practical use, an HSI-based CAD system using an interactive decision threshold is likely to be valuable. Experiments were also performed to measure the benefits of combining the colorectal and esophagogastric datasets (22 patients), and this yielded significantly better results with the MLP and SVM models.

Optical diagnosis of colorectal polyps using convolutional neural networks

Colonoscopy remains the gold standard investigation for colorectal cancer screening as it offers the opportunity to both detect and resect pre-malignant and neoplastic polyps. Although technologies for image-enhanced endoscopy are widely available, optical diagnosis has not been incorporated into routine clinical practice, mainly due to significant inter-operator variability. In recent years, there has been a growing number of studies demonstrating the potential of convolutional neural networks (CNN) to enhance optical diagnosis of polyps. Data suggest that the use of CNNs might mitigate the inter-operator variability amongst endoscopists, potentially enabling a “resect and discard“ or ”leave in“ strategy to be adopted in real-time. This would have significant financial benefits for healthcare systems, avoid unnecessary polypectomies of non-neoplastic polyps and improve the efficiency of colonoscopy. Here, we review advances in CNN for the optical diagnosis of colorectal polyps, current limitations and future directions.

Application of artificial intelligence in diagnosis of osteoporosis using medical images: a systematic review and meta-analysis

Artificial intelligence (AI) is a potentially reliable assistant in the diagnosis of osteoporosis. This meta-analysis aims to assess the diagnostic accuracy of the AI-based systems using medical images. We searched PubMed and Web of Science from inception to June 15, 2020, for eligible articles that applied AI approaches to diagnosing osteoporosis using medical images. Quality and bias of the included studies were evaluated with the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. The main outcome was the sensitivity and specificity of the performance of the AI-based systems. The data analysis utilized the R Foundation packages of "meta" for univariate analysis and Stata for bivariate analysis. Random effects model was utilized. Seven studies with 3186 patients were included in the meta-analysis. The overall risk of bias of the included studies was assessed as low. The pooled sensitivity was 0.96 (95% CI 0.93-1.00), and the pooled specificity was 0.95 (95% CI 0.91-0.99). However, high heterogeneity was found in this meta-analysis. The results supported that the AI-based systems had good accuracy in diagnosing osteoporosis. However, the high risk of bias in patient selection and high heterogeneity in the meta-analysis made the conclusion less convincing. The application of AI-based systems in osteoporosis diagnosis needs to be further confirmed by more prospective studies in multi-centers including more random samples from complete patient types.

Endoscopic imaging in inflammatory bowel disease: current developments and emerging strategies

INTRODUCTION

Developments in enhanced and magnified endoscopy have signified major advances in endoscopic imaging of ileocolonic pathology in inflammatory bowel disease (IBD). Artificial intelligence is increasingly being used to augment the benefits of these advanced techniques. Nevertheless, treatment of IBD patients is frustrated by high rates of non-response to therapy, while delayed detection and failures to detect neoplastic lesions impede successful surveillance. A possible solution is offered by molecular imaging, which adds functional imaging data to mucosal morphology assessment through visualizing biological parameters. Other label-free modalities enable visualization beyond the mucosal surface without the need of tracers.

AREAS COVERED

A literature search up to May 2020 was conducted in PubMed/MEDLINE in order to find relevant articles that involve the (pre-)clinical application of high-definition white light endoscopy, chromoendoscopy, artificial intelligence, confocal laser endomicroscopy, endocytoscopy, molecular imaging, optical coherence tomography, and Raman spectroscopy in IBD.

EXPERT OPINION

Enhanced and magnified endoscopy have enabled an improved assessment of the ileocolonic mucosa. Implementing molecular imaging in endoscopy could overcome the remaining clinical challenges by giving practitioners a real-time in vivo view of targeted biomarkers. Label-free modalities could help optimize the endoscopic assessment of mucosal healing and dysplasia detection in IBD patients.

Application of Artificial Intelligence in Medicine: An Overview

Artificial intelligence (AI) is a new technical discipline that uses computer technology to research and develop the theory, method, technique, and application system for the simulation, extension, and expansion of human intelligence. With the assistance of new AI technology, the traditional medical environment has changed a lot. For example, a patient's diagnosis based on radiological, pathological, endoscopic, ultrasonographic, and biochemical examinations has been effectively promoted with a higher accuracy and a lower human workload. The medical treatments during the perioperative period, including the preoperative preparation, surgical period, and postoperative recovery period, have been significantly enhanced with better surgical effects. In addition, AI technology has also played a crucial role in medical drug production, medical management, and medical education, taking them into a new direction. The purpose of this review is to introduce the application of AI in medicine and to provide an outlook of future trends.

Role of artificial intelligence in the diagnosis of oesophageal neoplasia: 2020 an endoscopic odyssey

The past decade has seen significant advances in endoscopic imaging and optical enhancements to aid early diagnosis. There is still a treatment gap due to the underdiagnosis of lesions of the oesophagus. Computer aided diagnosis may play an important role in the coming years in providing an adjunct to endoscopists in the early detection and diagnosis of early oesophageal cancers, therefore curative endoscopic therapy can be offered. Research in this area of artificial intelligence is expanding and the future looks promising. In this review article we will review current advances in artificial intelligence in the oesophagus and future directions for development.

Techniques to integrate artificial intelligence systems with medical information in gastroenterology

Gastrointestinal (GI) endoscopy is the central element in contemporary gastroenterology as it provides direct evidence to guide targeted therapy. To increase the accuracy of GI endoscopy and to reduce human-related errors, artificial intelligence (AI) has been applied in GI endoscopy, which has been proved to be effective in diagnosing and treating numerous diseases. Therefore, we review current research on the efficacy of AI-assisted GI endoscopy in order to assess its functions, advantages and how the design can be improved.

Surveillance of neo-squamous epithelium after ablation of Barrett's esophagus: is it better to use jumbo over standard biopsy forceps?

BACKGROUND AND AIMS

As obtaining adequate tissue on biopsy is critical for the detection of residual and recurrent intestinal metaplasia/dysplasia in Barrett's esophagus (BE) patients undergone Barrett's endoscopic eradication therapy (BET), we decided to compare the adequacy of biopsy specimens using jumbo versus standard biopsy forceps.

METHODS

This is a two-center study of patients' post-radiofrequency ablation of dysplastic BE. After BET, jumbo (Boston Scientific©, Radial Jaw 4, opening diameter 2.8 mm) or standard (Boston Scientific©, Radial Jaw 4, opening diameter 2.2 mm) biopsy forceps were utilized to obtain surveillance biopsies from the neo-squamous epithelium. Presence of lamina propria and proportion of squamous epithelium with partial or full thickness lamina propria was recorded by two experienced gastrointestinal pathologists who were blinded. Squamous epithelial biopsies that contained at least two-thirds of lamina propria were considered 'adequate'.

RESULTS

In a total of 211 biopsies from 55 BE patients, 145 biopsies (29 patients, 18 males, mean age 61 years, interquartile range [IQR] 33-83) were obtained using jumbo forceps and 66 biopsies (26 patients, all males, mean age 65 years, IQR 56-76) using standard forceps biopsies. Comparing jumbo versus standard forceps, the proportion of specimens with any subepithelial lamina propria was 51.7% versus 53%, P = 0.860 and the presence of adequate subepithelial lamina propria was 17.9% versus 9.1%, P = 0.096 respectively.

CONCLUSIONS

Use of jumbo forceps does not appear to have added advantage over standard forceps to obtain adequate biopsy specimens from the neo-squamous mucosa post-ablation.