Researchers at the University of San Francisco have developed an artificial intelligence capable of detecting Alzheimer’s disease in a patient six years before a doctor.
Scientists are very interested in the benefits of artificial intelligence in health. This time we are interested in the project of researchers from the University of San Francisco. They want to improve the diagnosis of Alzheimer’s disease. Yes, they do, According to the WHO, this disease affects no fewer than 35 million people worldwide, including 3 million in France.
To do this, Jae Ho Sohn, mentor of the Big Data team in radiology at UCSF, worked with the rest of the team to develop an artificial intelligence capable of diagnosing Alzheimer’s disease before a human doctor. He explains that the disease is difficult to spot in the early stages. “By the time all clinical symptoms are present and a definite diagnosis can be made, too many neurons have died, making the disease irreversible.“He explains in an article published by UCSF about him.
Detecting Alzheimer’s disease before human doctors do
In order to improve disease detection, the researchers therefore fed positron emission tomography into their machine learning algorithm. These are scans of the brain that allow three-dimensional measurement of brain activity (metabolism, blood flow, neuronal activity, glucose levels, etc.).). By monitoring glucose levels in the brain (especially in the frontal and parietal lobes), radiologists are able to indicate whether Alzheimer’s disease is on the horizon.
Because changes in the way cells assimilate glucose are subtle, researchers combined neuroimaging and machine learning to detect them. Trying to predict the onset of Alzheimer’s disease is a matter of timing. The test is done at the first signs of memory loss.
An AI with record accuracy
The algorithm was trained using 1921 scans and then tested on two databases. The first database consisted of 188 scans from a database known to artificial intelligence, but which had not yet been tested. The second database consisted of data from 40 patients who presented to the Memory and Aging Center at UCSF with cognitive impairment. In the first case, she identified 92 per cent of those with the disease. In the second case, the result rose to 98%. She was able to make this diagnosis on average 6 years before a human doctor: an impressive result.
If there’s no cure for Alzheimer’s, Detecting this degenerative disease in advance can provide a better end of life for patients. For Sohn and the rest of the team, this is a first step. They need to calibrate the artificial intelligence so that it can identify patients with the disease in larger cohorts and according to a wider range of criteria. A project to be followed up in the next 5 to 10 years.