Royal North Shore Hospital (RNSH) has launched a pioneering clinical trial to test whether artificial intelligence (AI) can match human expertise in a critical stage of cancer treatment – radiotherapy contouring. AI put to the test in radiotherapy treatment for breast cancer
Contouring is a process whereby a radiation oncologist meticulously outlines the tumours that will be targeted with radiation therapy and nearby organs at risk by hand on a CT scan. According to RNSH radiation oncology fellow Dr Joseph Chan, the process can take anywhere between 30 minutes to a few hours, depending on its complexity. He likens it to an architect creating a set of drawings that a builder will then use. “We make these sets of drawings, then technical staff help us plan the treatment,” he explains. While commercial AI software has been introduced to assist with contouring of normal body tissue organs in recent years, Joseph says there is no evidence that it performs as well as humans. “No clinical trial has proven that it’s actually equivalent to what a human does,” he says. “Can you imagine if a drug was being used and they never ran a randomised control trial?” To address this, Joseph is leading a trial involving the scans of 444 breast cancer patients undergoing radiotherapy to find out how AI contouring compares to radiation oncologists doing it by hand. The trial began in February 2025 and is expected to run over the next two years at RNSH, as well as Dubbo and Orange hospitals in rural NSW.
The team is comparing three contouring methods in the trial: commercial software, the RNSH-developed AI, and the contouring they do by hand. For every four patients, three will have AI-assisted contouring on the commercial and in-house systems, which will then be edited by radiation oncologists. One in four patients will have their scans contoured from scratch. The team has built their own AI for a few reasons. Firstly, with commercial software, there is no transparency around the data that was used to build the software. “We don’t know how high quality it is or where the data has come from,” says Joseph. The commercial software also takes a blanket approach. “They build these systems for everyone in the world – they sell it in Indonesia, Australia or America,” he says. “Your patient population in different centres can be quite different but these companies use a one-size-fits-all approach.” Due to current regulations, commercial software cannot be fine-tuned or re-trained on local patient data to better match local conditions. These limitations can reduce the potential gains from AI contouring. For example, if the commercial software always produces contours too large, then the radiation oncologist may have to spend more time editing than starting from scratch without AI. They also cannot teach this edit to the AI. “Every single time, it doesn’t save me time, and I’m having to adjust it at the same point,” he says.
The research team: Dr Marita Morgia, Dr Joseph Chan, A/Prof Sue Carroll, Dr Rob Finnegan and Dr Gabi Metz
28 Tomorrow’s medicine today
Made with FlippingBook - professional solution for displaying marketing and sales documents online