Although Alzheimer's disease affects tens of millions of people throughout the world, it is still difficult to detect at an early stage. But researchers who deal with the possibility of artificial intelligence in medicine have found that technology can help early diagnosis of dangerous diseases. The California team recently published a report on its study in Radiology magazine and showed how once a neural network is able to diagnose Alzheimer's disease in a limited number of patients based on visualization of brain imaging carried out years before the patient is actually diagnosed by a doctor.
This team uses brain imaging (FDG-PET imaging) to train and test their neural networks. In FDG, the patient's blood flow image is injected with a radioactive type of glucose, and then the body's tissues, including the brain, push it to the surface. Scientists and doctors can then use PET scans to sense the metabolic activity of this tissue, depending on how much FDG is taken.
The FDG-PET method is used to diagnose Alzheimer's, with patients who have a disease usually showing lower levels of metabolic activity in certain parts of the brain. Experts, however, must analyze these images to find evidence of disease, and this becomes very difficult because moderate cognitive impairment and Alzheimer's disease can cause the same results in scanning.
Therefore, the team used 2,109 FDG-PET images from 1002 patients, trained their neural networks at 90% and tested them on the remaining 10%. He also ran a test with a set of 40 patients scanned between 2006 and 2016, then compared artificial intelligence findings with a group of specialists who analyzed the same data.
With a series of separate test data, Artificial Intelligence is able to diagnose Alzheimer's patients with 100% accuracy and with an accuracy of 82% of those who do not suffer from dangerous diseases. He can also make an average estimate of more than the next six years. In comparison, the group of doctors who saw the same scanned image identified patients with Alzheimer's disease in 57% of cases and those without disease – at 91%. However, differences in machine and human performance are not very visible when it comes to diagnosing mild cognitive impairment that is not typical of Alzheimer's disease.
The researchers noted that their study had several limitations, including a small number of test data and limited types of training data.