Artificial Intelligence Promises More Timely Detection of Alzheimer’s Disease
One prominent obstacle that makes Alzheimer’s disease hard to treat is the inability to detect its occurrence early enough. But that story may be about to change for good with a new diagnostic method that enlists artificial intelligence (AI).
In a study published in the journal Radiology, researchers at the University of California, San Francisco (UCSF) were able to detect Alzheimer’s roughly six years earlier. They achieved this feat by combining neuroimaging with a machine-learning algorithm.
This finding could mean so much for people with Alzheimer’s, a condition that currently has no cure.
“One of the difficulties with Alzheimer’s disease is that by the time all the clinical symptoms manifest and we can make a definitive diagnosis, too many neurons have died, making it essentially irreversible,” said Jae Ho Sohn, MD, a resident at UCSF’s Department of Radiology and Biomedical Imaging.
Radiologists typically rely on brain scans in trying to diagnose this disorder. However, because it develops slowly, it is difficult to pin down the harmful changes that are taking place.
The high likelihood of missing subtle changes in Alzheimer’s is what makes it difficult to fight. There are drugs that could help slow its progression, however, making it crucial to be able to detect it early.
Brain scans for Alzheimer’s
Researchers have investigated positron emission tomography (PET) scans as helpful tools for timely diagnosis of Alzheimer’s. These imaging tests gauge the amounts of glucose or certain other molecules in the brain, such as proteins.
For the diagnosis of this disorder, PET scans typically look for glucose. These types are cheaper and are much more common, particularly in developing countries and smaller healthcare facilities.
PET scans for glucose are also useful for cancer staging.
Cells in the brain rely so much on glucose for energy production. Highly-active cells require more of this substance than their less-active counterparts.
However, cells use less glucose and, eventually, none as they become less efficient and die off.
Novel diagnostic technique
Radiologists usually check for glucose level reduction in the brain with scans when trying to diagnose Alzheimer’s. The focus of these investigations is, in particular, the frontal and parietal brain lobes.
However, changes in glucose levels while the disorder develops are slight and easy to miss. It, therefore, becomes tricky to detect early enough to inhibit its progression more effectively.
Sohn and colleagues decided to combine artificial intelligence with brain scans to see if it would improve the detection rates. The radiologist said the deep machine-learning algorithm used in the study was “particularly strong at finding very subtle but diffuse processes.”
The research team made use of images from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a large public database of brain scans, to train the algorithm. These images were from people that later had mild cognitive impairment, Alzheimer’s, or no disorder.
The algorithm was trained on a total of 1,921 PET scans. This enabled it to learn what changes could predict Alzheimer’s.
Researchers then proceeded to test the machine-learning algorithm on two datasets that were not previously used for training. One was a set of 188 images from the same ADNI database. The second comprised scans from 40 patients who possibly had cognitive impairment and made available to the UCSF Memory and Aging Center.
The algorithm performed very impressively on these scans. It was able to identify 92 percent of patients in the first dataset that developed Alzheimer’s. The detection rate was even higher in the second set at 98 percent.
What is next?
Perhaps, the more interesting part is that artificial intelligence predicted this disorder 75.8 months, on average, before clinical diagnosis was possible.
The researchers said their next plan is to test the algorithm on more varied, larger datasets.
“I believe this algorithm has the strong potential to be clinically relevant,” Sohn said. “However, before we can do that, we need to validate and calibrate the algorithm in a larger and more diverse patient cohort, ideally from different continents and various different types of settings.”
Neurologists will find this novel approach highly useful for treating patients if proven in larger studies. It could help patients with Alzheimer’s and mental health concerns to access treatments earlier.