Telltale Signs: Artificial intelligence can help to detect Alzheimer's disease years before the first symptoms appear. In a first pilot study, researchers trained a learning algorithm to identify subtle signs of dementia in special brain scans. These changes in brain metabolism are often hardly recognizable to physicians. The AI system, however, reached a hit rate of 100 percent, the researchers report.
© lowerek / thinkstock
It is one of the great assets of artificial intelligence: computer systems for learning are especially good at detecting subtle similarities in datasets. By means of training with extensive sample data, they can learn independently to identify for example colon cancer precursors, breast cancer or melanoma in medical images. In part, the hits of such machine brains are higher than those of experienced doctors.
Log in with the metabolism of the brain
But can the AI also interpret signs that even medical professionals usually miss? A classic case is early detection in dementia in Alzheimer's disease. Usually the progressive breakdown of brain matter is only diagnosed after the first symptoms manifest themselves. "But then the loss of brain matter is so severe that it is usually too late for interventions," said Jae-Ho Sohn of the University of California in San Francisco.
The problem: although there are subtle shifts in brain metabolism that can be visualized with a special version of positron emission tomography (PET). The uptake of radioactively labeled sugar molecules is assigned to different brain regions. However, the changes that are typical of the early stages of Alzheimer's are difficult to detect. "We humans are good at finding specific biomarkers for diseases," Sohn explains. "But these metabolic changes represent a more subtle and diffuse process."
Hit percentage 100 percent
Artificial intelligence comes into play at this point: Son and his team wanted to know whether the good pattern recognition of a learning algorithm could better detect the Alzheimer's sign in brain scan than a human being. For their experiment they trained the AI system at an early stage on a good 2,000 PET images of Alzheimer patients. Afterwards, the AI 40 would not have to assess the previously seen recordings.
The amazing result: the artificial intelligence had a hit percentage of 100 percent. "We were very satisfied: the system could identify every single case that later developed into Alzheimer's disease," says Sohn. Some of these PET images were taken several years before the onset of the first symptoms and a definitive diagnosis.
Early diagnosis – better therapy
According to the researchers, this proves that AI systems can improve the early detection of dementia. Although this was only a pilot study that needs to be revised and supplemented with more extensive multicenter studies. Nevertheless, Sohn and his team are convinced that adaptive algorithms can support radiologists and neurologists in the early detection of Alzheimer's disease in the future.
Early identification can also be helpful in finding new therapies: "The faster recognition of Alzheimer's disease offers researchers the opportunity to find new and better ways to stop or slow the progression of the disease," says son. (Radiology, 2018)