A new co-training AI algorithm for medical imaging has been developed by Monash University researchers, and it successfully simulates the process of getting a second opinion.

The study recently published in Nature Machine Intelligence used an adversarial, or competitive, learning approach against unlabelled data to address the dearth of human annotated, or labelled, medical images.

The engineering and IT faculties at Monash University are conducting research that will advance the field of medical image analysis for radiologists and other healthcare professionals.

The goal of the research design, according to PhD candidate Himashi Peiris of the Faculty of Engineering, was to establish competition between the two parts of a "dual-view" AI system.

According to Ms. Peiris, "one portion of the AI system attempts to mimic how radiologists read medical images by labelling them, while the other portion of the system judges the quality of the AI-generated labelled scans by benchmarking them against the sparsely labelled scans provided by radiologists."

Radiologists and other medical professionals traditionally hand-annotate or label medical scans to draw attention to particular regions of interest, such as tumours or other lesions. These labels offer direction or oversight for the training of AI models.

This method, which is time-consuming, error-prone, and causes patients to wait longer for treatments, depends on the subjective interpretation of specific individuals.

Since it takes a lot of work, time, and expertise to manually annotate numerous images, there are rarely many large-scale annotated medical image datasets available.

The algorithm created by the Monash researchers enables a variety of AI models to take advantage of the special benefits of labelled and unlabeled data, and learn from one another's predictions to increase accuracy overall.

According to Ms. Peiris, "Using a 10% labelled data setting across the three publicly available medical datasets, we achieved an average improvement of 3% compared to the most recent state-of-the-art approach under identical conditions."

"Our algorithm has outperformed previous state-of-the-art techniques in semi-supervised learning, achieving ground-breaking results. Unlike algorithms that depend on large amounts of annotated data, it exhibits remarkable performance even with few annotations.

As a result, AI models can uncover more accurate diagnoses and treatment options, validate their initial assessments, and make better decisions.

The research's next phase will extend the application's ability to work with various medical image types and create a specialised end-to-end product that radiologists can use in their daily practises.

💾💾💾

In the prestigious National Intelligence and Security Discovery Research Grants Programme, Monash University-led research into growing human brain cells onto silicon chips, with new continual learning capabilities to transform machine learning, has been awarded nearly $600,000.

In a dish, 800,000 brain cells are grown as part of a new research programme run by Associate Professor Adeel Razi of the Turner Institute for Brain and Mental Health in partnership with Melbourne start-up Cortical Labs. These brain cells are then "taught" to carry out tasks that have a clear goal. The team's research last year garnered widespread attention for the brain cells' capacity to play the straightforward computer game Pong, which is similar to tennis.

The research program's use of silicon chips embedded with lab-grown brain cells "merges the fields of artificial intelligence and synthetic biology to create programmable biological computing platforms," according to Associate Professor Razi.

Future versions of this new technology might eventually outperform currently-available, entirely silicon-based hardware in terms of performance.

"Such research's findings would have significant implications for a wide range of fields, including but not limited to planning, robotics, advanced automation, brain-machine interfaces, and drug discovery, giving Australia a significant strategic advantage."

Because the next generation of machine learning applications, including self-driving cars and trucks, autonomous drones, delivery robots, intelligent handheld and wearable devices, "will require a new type of machine intelligence that is able to learn throughout its lifetime," Associate Professor Razi said, the project received funding from the esteemed Australian grant body.

This "continual lifelong learning" enables machines to learn new skills without compromising their existing ones, adapt to changes, and use previously acquired knowledge to complete new tasks—all while sparing scarce resources like memory, energy, and processing power. This is impossible for modern AI, which also experiences "catastrophic forgetting."

Brains, on the other hand, are excellent at lifelong learning.

The goal of the project is to grow human brain cells in the DishBrain system, a laboratory dish, in order to comprehend the various biological mechanisms underlying lifelong continual learning.

"We plan to use this grant to build AI systems that can learn just like these biological neural networks do. This will enable us to scale up the hardware and method capacities until they can successfully replace in silico computing, according to Associate Professor Razi.

Cortical Labs' Chief Scientific Officer, Dr. Brett Kagan, applauded the funding. "The opportunity to harness the only material known to produce generalised intelligence - living neurones - offers an enormous potential to generate systems with capabilities not currently available, but can also provide insight into the most fundamental mechanism that grant us as humans our unique intelligence," he said.

The funding needed to carry out this cutting-edge, innovative research "offers the potential to both unlock new technologies and the possibility to start growing an Australian ecosystem in a blossoming new field," according to the statement.

Got a news tip for our journalists? Share it with us anonymously here.

Send press releases to [email protected]. Other ways to contact us. Editorially, we may rewrite headlines and descriptions.

Recommend Redwires AU: Accessible News For Young Cybersecurity Aussies

Redwires AU provides Young Australians with easily accessible, curated cybersecurity news.

Before you go..

You can get RedWires AU for free right now. Your donation, no matter how big or small, will help us keep doing honest journalism.

The readers of Redwires AU are the engine that drives our publication. Add your support to the effort to create a sustainable future for journalism that does not make compromises in the AU.

In the world we live in now, accurate and thorough reporting and analysis are becoming more and more critical. To stop the spread of false information, it's essential that everyone in Australia has access to good reporting.

The Redwires AU contributes to society by opening up access to information and resources for all people, rather than just a select few.

Our only goal is to educate the general public more thoroughly. If you believe in what we're trying to accomplish here, please consider making a contribution right away to ensure our success in the years to come.

Upgrade your subscription to get the most out of it. Join the growing number of people around the world who believe in the power of independent media.