Artificial intelligence (AI) is the future for upper GI endoscopy and is likely to improve the accurate detection and real-time diagnosis of premalignant and malignant lesions, Australian researchers predict.
However it may not necessarily require the intensive training and validation with huge, high-quality image databases that traditional AI algorithms require.
According to an article in the Journal of Gastroenterology and Hepatology, non-image-based AI platforms such as Raman spectroscopy might be a promising alternative.
The article, co-authored by Professor Rajvinder Singh and Dr Seon Ho Shin from the Lyell McEwin Hospital and University of Adelaide, said the technique is able to “capture biomolecular information when tissue molecules are agitated by a laser beam delivered through a fibre-optic probe.”
“Any cellular transformation, with its resultant changes in protein, DNA, and lipid content, will be reflected as changes in the spectral pattern. Cancerous and precancerous tissues have different molecular “fingerprints” from healthy tissue.”
A diagnosis, requiring no morphological information, can be displayed in real time.
“Preliminary data have demonstrated its ability to efficiently guide operators in performing targeted biopsies of dysplasia in Barrett’s oesophagus, gastric mapping for intestinal metaplasia, and in determining the margin of early upper GI cancer during endoscopic resection,” they said.
The article said the requirement for high-quality images had been the biggest limitation to image-based AI.
Nevertheless there was increasing evidence in gastric cancer, dysplasia in Barrett’s oesophagus and early oesophageal squamous cell cancer.
“The movement towards adoption of AI in endoscopy is inevitable as the technology has gained popularity in other industries,” the authors said.
“AI will likely improve the ability of general endoscopists for better detection and more accurate optical diagnosis, potentially closing the gap between generalists and experts. It also has the potential to also aid expert endoscopists in evaluating complex early upper GI cancers such as superficial oesophageal SCC, early neoplastic Barrett’s lesions and EGC.”
“An accurate AI system has the potential to be incorporated with a “live feedback tool,” which could then train endoscopists to recognise and characterise lesions.”
However the article noted the risk of relying on AI to the extent of potentially deskilling endoscopists.
“We must not train the next generations of endoscopists as technicians without the required capability to comprehend visual information. The role of AI must be to enhance cognition, not replace it,” they said.