eRSA services in use: Storage & Cloud
Pictures are worth a thousand words according to the old saying, but the images at the centre of Professor Lyle Palmer’s research are proving to be far more valuable.
The computer-based medical imaging that forms the cornerstone of Professor Palmer’s research in the new field of ‘radiomics’ is being used to predict death and probable diseases in people years in advance, enabling early intervention.
How does it work?
The research employs a concept called “deep learning”, an artificial intelligence technique that trains computers to see patterns in images that humans cannot. Deep learning is the technology that underlies self-driving cars, smart assistants such as Siri, and a range of stunning achievements in science and industry.
The system predicts patient outcomes by incorporating and analysing large volumes of image data to detect subtle patterns that doctors may miss.
“In CT images, the algorithm allowed us to explore even the most minute changes occurring in the tissues of the thorax that goes beyond what doctors can see and decipher,” said Professor Palmer, the Professor of Genetic Epidemiology at the University of Adelaide.
“Applying this technology to routinely collected medical images has the potential to predict the beginning of life-threatening diseases such as heart disease, cancer, and diabetes. Knowledge of such diseases early on will enable doctors to provide streamlined and tailor-made treatments and medications.
Professor Palmer said the more data you can feed into the machine’s algorithms, the more reliable its outputs. This means access to a large amount of data is an important element of the research, which the research team has been able to access via South Australia’s public hospital system.
Taking the work forward
Early research results have reinforced the significant potential of using computer-based medical imaging to aid doctors in the diagnosis and prognosis of important medical conditions.
In their most recent success, Professor Palmer’s team used deep learning to analyse pelvic x-rays in order to diagnose hip fractures. Their algorithm managed to outperform emergency department doctors, GPs, and radiologists in being able to accurately diagnose a hip fracture. “This is the first demonstration in radiology of ‘superhuman’ performance”, said Professor Palmer.
Currently, the team is about to begin work on mammograms from women screened by BreastScreen SA in order to improve the diagnosis of breast cancer.
“The issue with the current methods is that around 5% of the women screened are called back for further testing because of a perceived anomaly in their scans, most of whom will not actually have breast cancer. This is expensive and very anxiety-inducing for the individual concerned,” he said. “We hope to be able to make the initial testing more accurate, in order to streamline the process and make it more efficient and less stressful.”
Professor Palmer points out that the technology is meant to augment current medical services wherever possible, not replace them.
“This technology won’t replace doctors, but will be able to help doctors make better and faster decisions.”
The future of the technique
While this type of direct clinical diagnosis and prognosis hasn’t been used in medical services yet, Professor Palmer believes it will be a reality very soon.
In the United States, a 2017 study showed that a deep learning algorithm could read retinal photographs and diagnose eye disease as well as the best ophthalmologists.
Similarly, a group of dermatologists used deep learning technology in 2017 to analyse photographs of moles on people’s skin to predict and diagnose skin cancer as well as the best dermatologists.
“These demonstrations were in Ophthalmology and Dermatology, and ours are in Radiology. In each case, they showed the algorithm could diagnose conditions at least as well as best specialists,” he said.
The value of eRSA
One of the most significant challenges the team faces is securely storing and managing the massive amounts of data they collect and have access to.
“Just one medical image can be 10 gigabytes, and digital image collections can be enormous. A lot of physical storage is needed, and proper methods are required to manipulate that data,”
“Last year, for example, we received a data set from the National Institute of Health in the United States of serial thoracic CT images taken from 60,000 adults. These needed to be stored and analysed, something eRSA has enabled us to do.
“eRSA has been incredibly helpful and valuable for us. The last thing we wanted was to have to build something ourselves to store data. Having eRSA set up and manage our storage meant we not only had access to the latest infrastructure available but also enabled us, a small research team in South Australia, to compete with the world’s best in this exciting new area.”
Want to talk to Lyle about his research and the tools used?
e: firstname.lastname@example.org | p: 08 8313 2158