A way out of the Brexit morass?
09 May 2019 – 14:15 | No Comment

Brexit-bound Britain will participate in this month’s European Parliament (EP) election, unless UK prime minister, Theresa May, and opposition leader, Jeremy Corbyn, manage to push the thrice-rejected EU withdrawal agreement through the House of Commons …

Read the full story »
Health

Energy & Environment

Circular Economy

Climate Change

Security

Home » Cancer, Colorectal Cancer, Digestive Cancers, Interviews

Research suggests AI is faster at detecting cancer

Submitted by on 19 Feb 2019 – 16:17

A new endoscopic system powered by artificial intelligence was recently shown to automatically identify colorectal adenomas during colonoscopy. The system, developed in Japan, has recently been tested in one of the first prospective trials of AI-assisted endoscopy in a clinical setting, with the results presented at the 25th UEG Week in Barcelona, Spain.

Colonoscopic polypectomy for all adenomatous polyps is considered to contribute to the reduction of both the incidence and mortality of colorectal cancers. However, a substantial number of unnecessary polypectomies are carried out for non‐neoplastic polyps because of endoscopists’ misdiagnoses, resulting in considerable financial concerns.

A new endoscopic system powered by artificial intelligence was recently shown to automatically identify colorectal adenomas during colonoscopy. The system, developed in Japan, has recently been tested in one of the first prospective trials of AI-assisted endoscopy in a clinical setting, with the results presented at the 25th UEG Week in Barcelona, Spain.

The new computer-aided diagnostic system uses an endocytoscopic image — a 500-fold magnified view of a colorectal polyp – to analyse approximately 300 features of the polyp after applying narrow-band imaging (NBI) mode or staining with methylene blue. The system compares the features of each polyp against more than 30,000 endocytoscopic images that were used for machine learning, allowing it to predict the lesion pathology in less than a second.

Preliminary studies demonstrated the feasibility of using such a system to classify colorectal polyps; however, until today, no prospective studies have been reported.

Speaking to Government Gazette recently, Dr Yuichi Mori from Showa University in Yokohama, Japan, who involved 250 men and women in whom colorectal polyps had been detected using endocytoscopy, explained: “The most remarkable breakthrough with this system is that artificial intelligence enables real-time optical biopsy of colorectal polyps during colonoscopy, regardless of the endoscopists’ skill. This allows the complete resection of adenomatous polyps and prevents unnecessary polypectomy of non-neoplastic polyps.”

“We believe these results are acceptable for clinical application and our immediate goal is to obtain regulatory approval for the diagnostic system,” added Dr Mori.

Moving forwards, the research team is now undertaking a multicentre study for this purpose and the team are also working on developing an automatic polyp detection system. “Precise on-site identification of adenomas during colonoscopy contributes to the complete resection of neoplastic lesions,” said Dr Mori. “This is thought to decrease the risk of colorectal cancer and, ultimately, cancer-related death.”
Is AI the future of colorectal cancer detection?

Compared to the research on automatic pathological prediction of polyps/cancers, that of automatic detection of polyps/cancers is delayed because it needs more computer power and more learning material. Actually, we cannot find any prospective trial on this academic field, while we have more than five prospective studies in automatic pathological prediction of polyps. However, with the emergence of deep learning, more and more studies are reported nowadays, thus we can expect a great advance in the area of automated polyps/cancers detection in a couple of years.

The algorithm showed a very high accuracy rate according to your results — was this expected? Were you and your team surprised?
We were surprised at the performance of the machine learning at first, because AI was significantly superior to novice endoscopists in terms of diagnostic abilities.

Why is there so much hope that AI systems can diagnose cancer better than humans? What makes AI better at this role than humans?
In my view, AI has two major strengths compared with a physician. First of all, AI is definitely objective. It is well known that a physician sometimes makes different diagnosis even on the same endoscopic images, if these images are shown at different times or in different situations. However, AI always outputs the same diagnosis. The second strength is the number of endoscopic images which AI has learned. We have already trained the AI with more than 70,000 endoscopic images from over 2,000 colorectal polyps. Such a large number of polyps cannot be experienced in a routine practice for a physician.

What are the current limitations of AI in detecting and diagnosing cancer or other diseases? How can these limitations be overcome?

The biggest limitation of AI systems for colonoscopy is that there are only a few studies evaluating the usefulness of AI systems in a clinical setting. Most available studies are experimental ones, thus the real performance of AI in a clinical practice is unknown. Therefore, we have to keep it in mind that the real performance of AI systems might be worse than we thought. In this point of view, it is strongly required to conduct high-quality, prospective clinical trials to validate “real performance” of the AI system before its implementation into a clinical practice. As for our AI model, we have already published a total of five pre-clinical studies in which sensitivities and accuracies were almost over 90% in diagnosis of neoplastic lesions.

How can this help save the lives of patients?

Precise, on-site identification of adenomas during colonoscopy contributes to whole resection of neoplastic polyps. Resection of all neoplastic polyps is believed to decrease both the future incidence of colorectal cancers and cancer-related death, which is a big benefit for patients.
What kind of savings could such a technological advancement bring to the cost of colorectal cancer screening?

We believe that this AI system offers good cost-effectiveness by avoiding some of the cost for polypectomy and subsequent pathological examinations. If the AI system correctly identifies neoplastic polyps requiring resection from non-neoplastic polyps not requiring resection, a large amount of costs related to unnecessary polypectomy for non-neoplastic polyps can be saved (eg, the cost for polypectomy and pathological assessment is approximately US$400 under Japanese National Health Insurance). This can be a good financial benefit for a colorectal cancer screening programme.