eHealthNews.nz: AI & Analytics

AI-enhanced digital pathology project to improve gastrointestinal cancer outcomes

Wednesday, 3 September 2025  

NEWS - eHealthNews.nz editor Rebecca McBeth 

A $700,000 AI in Health research grant will help to develop a tool that could halve unnecessary operations for early-stage colorectal cancer patients, researchers say.

Tim Eglinton and Arthur Morley-Bunker are part of a team that have secured a $700,000 grant from the Health Research Council to develop an artificial intelligence (AI) algorithm that could reduce unnecessary cancer surgeries.

The two-year project uses computer vision technology to analyse digital pathology slides and predict outcomes for patients with gastrointestinal cancers.

Eglinton says that up to 90 percent of early colorectal cancer patients undergo major bowel surgery they may not need, after having a cancer removed at colonoscopy.

"We are really bad at predicting whether those lymph glands next to the bowel are going to be involved or whether there is going to be residual cancer in the bowel if we go and do the operation," he tells eHealthNews.

Once the full surgery is done and the removed tissue is examined, they find no remaining cancer in up to 90 percent of cases.

The AI algorithm analyses digitised biopsy slides from over 800 patients across six major New Zealand hospitals. It is learning to identify patterns that predict whether patients will have residual cancer requiring surgery.

"The algorithm is picking up features in these slides that the human pathologists may not," Eglinton says. 

"It is looking at that whole pattern on the slide and saying, yes, this patient is going to have lymph node metastases."



Initial testing on 140 Canterbury patients has the algorithm performing at about 80 percent accuracy. Researchers expect this to improve significantly when the full dataset of 850 patients is used, hopefully reaching accuracy levels in the 90s when combined with traditional clinical factors.

The algorithm generates a heat map showing which areas of the biopsy slide are most predictive.

"It does not just say yes or no, it actually tells us the areas that are most predictive on the slide," he explains. 

"Pathologists can go back and look at that slide and compare and see that this is what the algorithm was looking at."

The second cancer being focused on is pancreatic cancer, because patients often undergo Whipple operations, which involve removing the pancreas and duodenum in a five or six-hour procedure with high mortality and morbidity rates.

"There is a huge number of patients that are going through major surgery that they are not going to benefit from," says Eglinton.

He says the close network of about 50 to 60 colorectal surgery specialists across the country has enabled researchers to gather comprehensive local datasets, with the aim of avoiding any bias from international datasets.

Morley-Bunker says the AI research builds on earlier work funded by the Cancer Society and the Colorectal Surgical Society of Australia and New Zealand. 

The project includes plans for clinical implementation, starting with a pilot programme in Canterbury. The AI system will generate user-friendly pathology reports with heat maps and percentage probabilities to help surgeons make treatment recommendations.

Eglinton adds that the algorithm could also prevent under-treatment as some patients choose surveillance over surgery and miss their opportunity for a cure.

The researchers hope to reduce unnecessary surgery by half in early colorectal cancer patients and say the methodology could potentially expand to other cancer types, such as breast cancer.

Image: Heat map showing areas of interest


To comment on or discuss this news story, go to the eHealthNews category on the HiNZ eHealth Forum


You’ve read this article for free, but good journalism takes time and resource to produce. Please consider supporting eHealthNews by becoming a member of HiNZ, for just $17 a month

 

Read more AI & Analytics news


Return to eHealthNews.nz home page