Associate Professor Zhen He is harnessing artificial intelligence (AI) to tackle some of healthcare’s biggest challenges, including how cancer is diagnosed. A computer scientist with a background in applied research, Assoc Prof He leads a deep-learning research group at La Trobe University focused on building practical, real-world systems with impact.
Deep learning is a branch of AI that involves training computer systems to recognise patterns in large datasets, helping them make predictions and solve complex problems. One example of this work is in digital pathology – the process of examining tissue samples via high-resolution images instead of glass slides under a microscope. Assoc Prof He’s team has developed AI models to help detect and count tumour-infiltrating lymphocytes – immune cells that cluster around cancer cells and are linked to better outcomes.
“A human simply can’t count them all – they focus on small areas and make rough estimates,” he says. “But with AI, you can literally count every single one.”
This not only reduces the workload for pathologists but also improves consistency.
“There’s less variability between observers, which means we can make more reliable predictions about patient outcomes,” Assoc Prof He says.
Working with Melbourne’s Austin Hospital, Assoc Prof He’s team has also addressed a common challenge in medical AI: ensuring their models work across data from different sources.
“Images from different scanners can vary slightly in colour or quality,” he says. “But we’ve shown that our method works across multiple independent cohorts, which is really important.”
Deep learning: From oncology to athletics
Zhen’s journey into AI began about a decade ago when he watched a video lecture by Andrew Ng – a global leader in machine learning and a co-founder of the deep-learning research team, Google Brain.
“I was really inspired by his presentation,” Assoc Prof He recalls. “So I told my research group, ‘I think this is probably a lot more fun than optimising databases’.”
Assoc Prof He and his team at La Trobe University shifted their focus to deep learning and computer vision at a time when the field was still emerging. “At first, we only understood about 20 or 30 percent of the papers we read – but eventually, we were up to 80 percent,” he says.
Their early move into the space gave them a valuable head start, leading to collaborations across several industries.
One of the most impactful was with the Australian Institute of Sport and Swimming Australia, where Assoc Prof He’s team developed SPARTA 2, a deep-learning system for swimming competition analysis. Traditionally, teams of analysts would manually tag video footage frame by frame, marking when each swimmer’s hand entered the water to count strokes and compare actual race execution to pre-set race plans.
“They’d have many analysts manually annotating swimmers late into the night during Australian Swimming Championships,” Assoc Prof He says. “We built an AI solution that ran on just two high-powered laptops, tracked all 10 lanes, and produced reports in near real time – more accurately than humans.”
SPARTA 2 was used for all swimming competition analysis at the Tokyo Olympics in 2021, where it operated flawlessly.
In another project, Assoc Prof He’s team created an AI tool to assist Telstra customer service agents in live chat sessions – the tool is now used by more than 1,500 agents.
Looking to the future
Today, Assoc Prof He’s team continues to apply deep learning to a range of real-world problems – from developing a voice-based AI tool to detect alcohol intoxication as a low-cost alternative to breathalysers, to analysing satellite imagery to track how land is used and how it’s changing in agriculture.
That breadth requires constant learning. “In every new area, I’ve had to build up domain knowledge from scratch – sometimes that means revisiting high school biology, sometimes it’s learning a whole new set of concepts and language,” Assoc Prof He says.
On the healthcare front, his group is also working on automating analysis of complex tumour datasets that can take months to interpret manually. “There’s a device that counts every cell in a tumour sample – about 600,000 cells – and identifies their type,” he explains. “But analysing the output takes people months. We’re building AI tools to do it much faster.”
He’s also exploring how AI can accelerate drug discovery.
“In a lab, testing one idea can take weeks,” Assoc Prof He says. “But with AI, you can try out millions of ideas in just a few days. A pathologist visited our lab and in one morning, we tested 20 different hypotheses – something that would have taken months otherwise.” Among the possibilities: discovering alternatives to antibiotics, or identifying new biomarkers to personalise cancer treatment.
When choosing projects, Assoc Prof He looks for good data and committed partners.
“Without data, you can’t do anything,” he says. “And the people – they have to be just as invested. They’re the ones who teach me the domain knowledge and help assemble the data. It has to be a collaboration.”
Assoc Prof He currently supervises five research students and says close collaboration is central to their approach. “One of my students has been with me for 10 years – from the beginning of my deep-learning journey. He knows as much as I do and is incredibly creative. Tapping into that kind of experience makes a huge difference.”
An appetite for discovery, Assoc Prof He says, is what keeps him motivated. “The idea of finding a new drug or a biomarker that helps oncologists make better decisions, that’s what really excites me,” he says.
Connect with Associate Professor Zhen He
La Trobe Profile: Zhen He
Email: Z.He@latrobe.edu.au
Assoc Prof He is presenting at the upcoming La Trobe Industy webinar event, 'ACAMI: Advancing Medical Frontiers with AI' on Wednesday 30 July.
To stay up to date on future La Trobe Industry events, subscribe to the Industry newsletter or Follow ‘La Trobe Industry’ on LinkedIn and Eventbrite.
