Updated: Jul 27, 2022
Colorectal cancer contributes to a substantial proportion of the global burden of morbidity and mortality. It is the third most common cancer, preventable by regular surveillance examinations during colonoscopy.
Colonoscopy is a critical screening, diagnostic and treatment tool, capable of interrupting adenoma to carcinoma progression. But being largely operator-dependent, the experience of the endoscopists influences the chances of having an optimal examination done. This is an important problem that results in the increased occurrence of interval cancers (resulting from failing to detect or completely resect carcinomas at colonoscopy).
AI creates the opportunity to standardize examinations and improve diagnostic and, subsequent, treatment outcomes.
How exactly does AI do this?
By sequentially and progressively accessing features of the images or video captured during a colonoscopy, AI systems increase the discernment of polyps and may proffer a start at the underlying histology. These features depend on the type of AI system used. Generally, it could be a computer-assisted polyp detection or classification system. The technology behind this is image technology integrated with machine learning.
Another advantage AI brings to colonoscopy is the reduction of unnecessary surgical recessions. By precisely examining the depth of lesions, superficial lesions can be resected without the added risks of surgery and added time and financial costs.
In spite of these advantages, there are concerns trailing its use and limiting broader application. A wide database of normal and abnormal colonoscopic images and or videos from a diverse pool of people needs curating in order to broaden the evaluation models used by AI systems. Also, annotated data to clarify ambiguities is constantly needed. This requires repeated expert assessment and agreement, which is not available at the level to ensure its reliability and widespread use. Since the form AI in colonoscopy takes is machine learning, it is understandable how these challenges pose limitations. Essentially, the system needs to learn from reliable data for effective functioning. This is clearly a challenge endoscopists and AI researchers would continue to improve on for its further development.
AI improves the screening of, and decisions to resect colon adenomas and carcinomas. Although it is an imperfect innovation, requiring input from multiple experts and diverse data, it is constantly evolving. Ultimately, it is capable of reducing the incidence of colorectal cancers on a wider scale than it does presently.
To learn more about AI in colonoscopy, the following resources will be a great place to start: