Although Alzheimer's disease affects tens of millions of people around the world, it is still difficult to detect it at an early stage. But scientists who deal with the possibilities of artificial intelligence in medicine have found that technology can help to diagnose treacherous diseases early. The California team has recently published a report on its Radiology study and demonstrated that once it was trained, the neural network was able to accurately diagnose Alzheimer's disease in a limited number of patients based on brain imaging visualizations that were performed years before the patients concerned were diagnosed by a physician.
The team uses imaging brain (FDG-PET imaging) to train and test their neural network. In the FDG, images of the patient's bloodstream are injected with the radioactive type of glucose, and his body tissue, including the brain, pushes him to the surface. Researchers and physicians can then use the PET scan to determine the metabolic activity of this tissue, depending on how much FDG is used.
The FDG-PET method is used to diagnose Alzheimer's disease, in patients who have a disease usually exhibit lower levels of metabolic activity in certain parts of the brain. However, experts have to analyze these images to find evidence of the disease, which becomes very difficult because mild cognitive impairment and Alzheimer's disease can lead to similar scan results.
Therefore, the team uses 2,109 FDG-PET images from 1002 patients, trains their neural network to 90% and tests them for the remaining 10%. She also runs tests with a single set of 40 patients scanned between 2006 and 2016, then compares findings of artificial intelligence with the results of a group of experts who analyze the same data.
With a separate set of test data, artificial intelligence is able to diagnose patients with Alzheimer's disease with 100% accuracy and 82% accuracy to those who do not suffer from a traumatic illness. It can also predict on average more than six years. For comparison, a group of physicians who participated in the same scanned images identified patients with Alzheimer's disease in 57% of cases and patients without disease – 91%. However, the differences in mechanical and human performance are not so noticeable as regards the diagnosis of mild cognitive impairment not typical of Alzheimer's disease.
Researchers note that their research has several limitations, including a small amount of trial data and limited types of training data.