New software built in Japan can detect bowel cancer in less than a second, researchers claim.
In recently-conducted trials, the artificial intelligence (AI)-powered system was able to spot colorectal adenomas — which are benign tumours that can evolve into cancer — from magnified endoscopic images. The images were matched against 30,000 others that were used for machine learning.
The system analysed more than 300 colorectal adenomas in 250 patients, taking less than a second to assess each magnified endoscopic image and determine the malignancy of the tumours with 94 percent accuracy, researchers claim.
“The most remarkable breakthrough with this system is that AI enables real-time optical biopsy of colorectal polyps during colonoscopy, regardless of the endoscopists’ skill,” said study leader Dr Yuichi Mori from Showa University in Yokohama, Japan, who presented the results at United European Gastroenterology Week in Barcelona, Spain.
While the system is yet to obtain regulatory approval, Mori believes it could spare many patients from needless surgery.
“This allows the complete resection of adenomatous (cancerous) polyps and prevents unnecessary polypectomy (removal) of non-neoplastic polyps,” he said.
“We believe these results are acceptable for clinical application and our immediate goal is to obtain regulatory approval for the diagnostic system.”
The early detection of cancer through the use of AI and other technologies is being explored globally. Earlier this year, the UK’s National Health Service (NHS) and Intel said they are working together to make cancer detection more efficient through artificial intelligence (AI). A team of scientists, hosted by the University of Warwick’s Tissue Image Analytics laboratory, have been creating a digital repository of known tumour and immune cells based on thousands of human tissue cells.
The database of cancer information will then be used by algorithms to recognise these cells automatically, the companies said at the time, with the collaboration initially focusing on lung cancer.
Architecture including TensorFlow, originally designed by Google, will form the basis for the AI system and will be powered by Intel Xeon processors, Intel and NHS said in May.
Also in May, the Commonwealth Scientific and Industrial Research Organisation (CSIRO) in Australia announced that researchers from its Data61 division have been developing software that could enable the detection of angiogenesis — the development of new blood vessels — which precedes the growth of cancers.
Researchers at Data61 and the Shanghai Institute of Applied Physics at the Chinese Academy of Sciences analysed 26 high-resolution 3D micro-CT images of the brains and livers of 26 mice at various stages of cancer growth and developed an algorithm that generates what CSIRO claims is an accurate geometric representation of blood vessels.
Data61’s software will enable the monitoring of subtle proliferations in blood vessels over time, providing a better understanding of how patients are responding to anti-angiogenesis treatment.
“The accurate quantification of vasculature changes, particularly the number of terminal vessel branches, can play a critical role in accurate assessment and treatment,” CSIRO said at the time.
This is not the first time CSIRO has looked into cancer detection; in December, the government-backed organisation announced that a new, more accurate blood test to detect bowel cancer recurrence, known as Colvera, had been launched in the United States.
The blood test, which detects cancer-specific chemical changes in fragments of DNA from tumours found circulating in blood, is the result of a collaboration between CSIRO, Flinders University, and Clinical Genomics.
California-based Guardant Health is also looking into early cancer detection and raised $360 million this year to sequence the tumour DNA of more than 1 million cancer patients within five years and use the data to develop blood-based tests for early cancer detection.
Freenome, a South San Francisco-based liquid biopsy startup that uses a combination of machine learning and biology to detect the cell-free DNA sequencing of cancer before it becomes deadly, also raised $71.2 million to date from investors such as Google Ventures, which has also backed Freenome competitor Grail.
In January, Grail, a spin-off of the NASDAQ-listed DNA-sequencing giant Illumina, raised $900 million in a Series B round led by Arch Venture Partners, with contribution from other investors such as Johnson and Johnson’s innovation arm.