Artificial Intelligence in Gastrointestinal Endoscopy: General Overview

Artificial Intelligence in Gastrointestinal Endoscopy: General Overview

Introduction

Artificial intelligence (AI) has become a trending topic in clinical medicine, particularly in the field of gastrointestinal (GI) endoscopy. The integration of AI into GI endoscopy holds the potential to significantly enhance the quality of endoscopic procedures by improving accuracy, consistency, and efficiency. AI can compensate for human errors and limitations, making endoscopic procedures more reliable and of higher quality. While AI has demonstrated promising results in both diagnostic and therapeutic endoscopy across various parts of the GI tract, further studies are required before its widespread adoption in clinical practice. Additionally, ethical considerations and new legislation may be necessary to govern its use. In conclusion, AI is poised to revolutionize GI endoscopy in the coming years, offering substantial improvements at all levels.

The Role of AI in Gastrointestinal Endoscopy

AI technology is designed to augment human capabilities by providing machines with reliable autonomy, increasing work productivity, and enhancing efficiency. In the context of GI endoscopy, AI can reduce inter-operator variability, improve diagnostic accuracy, and facilitate rapid and accurate therapeutic decision-making. Furthermore, AI has the potential to reduce the time, cost, and burden associated with endoscopic procedures.

AI-assisted endoscopy relies on computer algorithms that mimic human brain functions. These algorithms react to input information and leverage learned data to produce outputs. The core principle behind AI technology is machine learning (ML), which involves teaching computer algorithms to recognize patterns in data. ML enables algorithms to automatically learn and improve from experience without explicit programming. One of the most rapidly advancing ML methods is deep learning (DL), which uses multi-layered artificial neural networks inspired by the biological neural networks of the human brain. DL models can analyze data logically, identify patterns, draw conclusions, and make decisions, making them far more capable than standard ML models.

AI in Esophageal Endoscopy

AI has shown significant promise in improving esophageal cancer screening, particularly in the identification of dysplasia and cancer in Barrett’s esophagus (BE) and squamous cell carcinoma. The incidence of esophageal adenocarcinoma (EAC) has risen rapidly over the past four decades, largely due to increasing rates of obesity. EAC is often diagnosed at an advanced stage, resulting in a poor prognosis. Early detection of neoplastic changes in BE is crucial, especially given the availability of highly curative endoscopic treatments such as endoscopic mucosal resection and radiofrequency ablation.

Current screening methods for EAC involve direct endoscopic visualization coupled with guided or random biopsies. However, random biopsies, as per the Seattle protocol, are relatively inefficient, time-consuming, and yield a low diagnostic rate. AI-assisted endoscopy can improve the sensitivity and speed of EAC screening, reducing the burden on endoscopists who face the challenge of not missing early cancers.

Several studies have demonstrated the utility of AI in detecting EAC. For instance, Swager et al. developed a computer algorithm based on volumetric laser endomicroscopy (VLE) images to identify early BE neoplasia. The algorithm showed good performance in detecting BE neoplasia, with an area under the curve (AUC) of 0.95. Similarly, van der Sommen et al. tested a computer algorithm for detecting early neoplastic lesions in BE, achieving a sensitivity and specificity of 0.83 on a per-image analysis. Other studies, such as those by Horie et al., Shin et al., and Quang et al., have also demonstrated the effectiveness of AI in diagnosing esophageal cancer with high sensitivity and specificity.

AI in Gastric Endoscopy

AI offers invaluable assistance in the management of early gastric cancer (EGC) at various levels, including diagnosis, cancer staging, lesion delineation, and prediction of Helicobacter pylori (H. pylori) infection. Gastric adenocarcinoma is the third leading cause of global cancer mortality, with early detection and prompt treatment being critical for improving patient survival. However, EGC can be challenging to detect due to its nonspecific abnormalities, often leading to missed diagnoses during endoscopy.

Current screening for EGC relies on direct visualization during gastroscopy, aided by image-enhancing tools such as chromoendoscopy, narrow-band imaging (NBI), and magnification. AI can help reduce inter-operator variability and improve the detection of EGC-related mucosal abnormalities. Several studies have demonstrated the effectiveness of AI in diagnosing EGC. For example, Miyaki et al. developed a computer recognition system called the support vector machine (SVM)-based analysis system, which showed high accuracy in differentiating between EGC and surrounding tissue. Hirasawa et al. developed a deep learning-based system for EGC diagnosis, achieving a sensitivity of 92.2%. Other studies, such as those by Kubota et al. and Zhu et al., have also shown promising results in the diagnosis and staging of EGC.

AI in Wireless Video Capsule Endoscopy (VCE)

Wireless video capsule endoscopy (VCE) is recommended as the first-line diagnostic exam for small bowel (SB) exploration. While VCE is well-tolerated by patients, the analysis of the large amount of data it generates can be time-consuming and burdensome. AI can assist clinicians by automating the diagnosis of various conditions, including bleeding angioectasia, celiac disease, and intestinal hookworms.

AI has already proven effective in detecting SB bleeding. Early CADe systems for bleeding detection used color-based feature extraction to distinguish bleeding-containing frames from those without bleeding. More advanced systems combine color and texture descriptors for more accurate bleeding detection. For example, Hassan et al. developed a CAD system based on deep learning that achieved sensitivities and specificities as high as 99% for GI bleeding detection. Similarly, Xiao et al. developed a CAD system using deep learning that achieved a performance score of >99% for GI bleeding detection in VCE.

AI in Colonoscopy

AI has the potential to significantly improve colonoscopy by assisting in polyp detection, characterization, and classification, as well as in predicting mucosal inflammatory activity in inflammatory bowel disease (IBD) patients. Colorectal cancer (CRC) is the third most diagnosed cancer and the second leading cause of cancer-related deaths worldwide. Colonoscopy is an effective method for CRC screening and prevention, with the detection and complete resection of neoplastic polyps being critical for reducing CRC incidence.

However, the adenoma miss rate remains high, despite advances in technology and devices. AI can help reduce the miss rate by providing automated detection of colorectal polyps. Several CADe systems have been developed and tested for polyp detection. For example, Misawa et al. developed a CADe system that detected 94% of test polyps with a false-positive rate of 60%. Other studies, such as those by Fernández-Esparrach et al., Urban et al., and Wang et al., have also demonstrated the effectiveness of AI in polyp detection.

AI can also assist in the real-time optical biopsy of small polyps, reducing the need for unnecessary polypectomies of non-neoplastic polyps. Misawa et al. developed a CADx system based on endocytoscopy (EC) that provided highly accurate diagnoses in real-time. Other CADx systems, such as those developed by Gross et al., Kominami et al., and Chen et al., have also shown high accuracy in polyp characterization and classification.

Future Perspectives

AI is expected to become widely available in the near future, with many AI systems already demonstrating high accuracy for their intended tasks. However, some functions, particularly those related to cancer diagnosis and treatment, require further testing before they can be adopted into clinical practice. CADe and CADx systems for colon polyp detection and characterization are among the most promising AI applications that may soon be introduced into clinical practice.

The future of GI endoscopy will likely see a dramatic transformation with the integration of AI. While AI will initially serve as an assistant to endoscopists, it may eventually take on a more central role in decision-making. However, this raises important ethical and legal questions regarding responsibility for medical errors and machine malfunctions. Further research and regulations are needed to address these concerns and ensure the safe and ethical use of AI in GI endoscopy.

Conclusion

The application of AI in GI endoscopy has the potential to significantly improve the quality of endoscopic procedures at all levels. AI can reduce inter-operator variability, compensate for human errors, and enhance the accuracy and efficiency of endoscopic procedures. While further research and regulations are needed, AI is poised to revolutionize GI endoscopy in the coming years, offering substantial benefits for both patients and physicians.

doi.org/10.1097/CM9.0000000000000623

Was this helpful?

YesNo

0 / 0