Subscribe to RSS
DOI: 10.1055/a-1471-3474
Artificial intelligence for gastric cancer: can we make further progress?
Referring to Wu L et al. p. 1199–1207![](https://www.thieme-connect.de/media/endoscopy/202112/lookinside/thumbnails/20696_10-1055-a-1471-3474-1.jpg)
In this issue of Endoscopy, Wu et al. report on a real-time artificial intelligence (AI) system that detects early gastric cancer (EGC) and blind spots (areas that were not observed) [1]. The authors first developed a novel system using a deep convolutional neural network to analyze esophagogastroduodenoscopy images of EGC [2]. Thereafter, they refined the existing system by integrating the trained real-time EGC detection model, and called the system ENDOANGEL. ENDOANGEL improved the quality of endoscopic examinations by monitoring blind spots in real time.
“In endoscopy, AI with “true intelligence” needs to incorporate the decision-making processes of endoscopists, who do not base their judgments solely on color or shape when observing the gastric mucosa.”
The development of ENDOANGEL has the potential to improve endoscopic examination of the stomach. The stomach is an organ with a wide, curved lumen, in which certain blind spots may not be detectable even when the entire stomach is examined [3]. If AI can recognize these anatomical regions, the entire examination of the stomach could be assured. In addition, ENDOANGEL could be an effective training tool for unskilled endoscopists who seek to improve their observational abilities. Therefore, there is great potential for positioning AI as a tool that supports high quality endoscopy training.
Gastric cancer occurs in patients with chronic gastritis related to Helicobacter pylori infection. EGCs can resemble benign lesions (e. g. gastritis, erosions, ulcers) and are difficult to detect [4]. During real-time examination in the study by Wu et al., ENDOANGEL predicted the two advanced gastric cancers and the three EGCs that were confirmed by pathology, suggesting that a more detailed evaluation of this technique is warranted. In contrast, when ENDOANGEL examined endoscopic imaging, 2107 red boxes or concerning areas in 498 patients were highlighted as suspicious for gastric cancer. However, the authors determined that 1750 of these frames were due to noise caused by reflections, bubbles, and mucus. Biopsies were taken from the 196 lesions in the remaining 357 red boxes at the discretion of the endoscopists. While this number appears large, the number of lesions requiring histological confirmation was actually reduced by about half. This is a meaningful result and demonstrates the potential for the ENDOANGEL system to assist endoscopic detection of EGC.
AI has progressed significantly in line with technological innovations, such as deep learning and machine learning, which has excelled in its image-recognition ability. Currently, it is being used in the fields of genomic medicine, drug discovery [5], and radiology [6]. AI is rapidly expanding in the field of endoscopy, too. Initial work on AI systems for endoscopy has focused on the detection and optical diagnosis of colorectal polyps. Byrne et al. validated their AI model for polyp diagnosis using unaltered videos containing low quality image frames as a test set. They confirmed that the model provided over 90 % accuracy for differentiating between adenomatous and hyperplastic polyps [7].
However, AI in endoscopy is still in the realm of face-recognition systems; in other words, the current function of AI is to recognize images that are similar to previously memorized images. Therefore, the AI system performs a narrow task, similarly to the face-recognition system of an urban surveillance camera that searches for a criminal. What then is the meaning of “intelligence”? It is an ability to collect information and make qualified decisions based on this information. Intelligence goes beyond simply memorizing vast quantities of information; it also uses that information to make comprehensive decisions. From this perspective, the AI systems developed to date in some fields may not yet have attained true intelligence.
In any case, the gastric mucosa is sensitive to the inflammatory effects of gastric acid, bile acid, and the effects of diet. In terms of the shape, the diagnosis is not constant because the stomach is an organ with a wide, curved lumen rather than a tubular structure like the colon, rectum, and esophagus. Therefore, the form of cancer we should look for depends on various factors, and gastric examination requires a thought process from endoscopists that is more complex than that required when looking for colorectal polyps during a colorectal examination.
In endoscopy, AI with “true intelligence” needs to incorporate the decision-making processes of endoscopists, who do not base their judgments solely on color or shape when observing the gastric mucosa. Additionally, the process leading to a diagnosis is different for each endoscopist. Every specialist uses their own internal memory database of images from past experience and training, in association with various other factors. In addition, the information storage and recall processes change rapidly and unknowingly according to the situation. Thus, education for endoscopic diagnosis is difficult because it is an unconscious process [8]. In addition, the endoscopist’s internal database is updated daily. With the help of more advanced AI systems that can “talk” to humans, such as HAL 9000, the AI that appeared in the movie “2001: A Space Odyssey”, we may be able to elucidate the unconscious diagnostic processes that are not well understood.
Currently, the optimal way to interact with AI is to build a cooperative relationship between AI and humans, where the AI technology supports human decision making. The medical perspective on the AI era is to improve existing human capability and to aim for a collaboration with AI systems, which are expected to evolve.
The development of HAL 9000-type technology may be on the horizon, and I believe that the appearance of such a system in the field of endoscopy will mark the emergence of “true intelligence.” Of course, humans should not give contradictory commands, such as those that caused HAL 9000’s failures; similarly, the appearance of Skynet (the AI that appeared in the movie “Terminator”), which has an ego, is not desirable! As a general AI system evolves, the question of the role of humans will become increasingly important. While we await the emergence of a general AI with comparable thought processes to human endoscopists, we should discuss how we want to shape our interaction with AI in the near future. The study by Wu et al. is the beginning of this process.
Publication History
Article published online:
27 July 2021
© 2021. Thieme. All rights reserved.
Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany
-
References
- 1 Wu L, He X, Liu M. et al. 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. Endoscopy 2021; 53: 1199-1207
- 2 Wu L, Zhou W, Wan X. et al. A deep neural network improves endoscopic detection of early gastric cancer without blind spots. Endoscopy 2019; 51: 522-531
- 3 Yao K, Uedo N, Kamada T. et al. Guidelines for endoscopic diagnosis of early gastric cancer. Dig Endosc 2020; 32: 663-698
- 4 Yalmarthi S, Witherspoon P, McCole D. et al. Missed diagnoses in patients with upper gastrointestinal cancers. Endoscopy 2004; 36: 874-879
- 5 Stokes JM, Yang K, Swanson K. et al. A deep learning approach to antibiotic discovery. Cell 2020; 180: 688-702
- 6 Murphy K, Smith H, Knoops AJG. et al. COVID-19 on chest radiographs: a multireader evaluation of an artificial intelligence system. Radiology 2020; 296: E166-E172
- 7 Byrne MF, Chapados N, Soudan F. et al. Real-time differentiation of adenomatous and hyperplastic diminutive colorectal polyps during analysis of unaltered videos of standard colonoscopy using a deep learning model. Gut 2019; 68: 94-100
- 8 Biederman I, Shiffrar MM. Sexing day-old chicks: a case study and expert systems analysis of a difficult perceptual-learning. J Exp Psychol Learn Mem Cogn 1987; 13: 640-645