AI can predict Alzheimer six years before diagnosis

April 01, 2019

Alzheimer’s disease affects tens of millions of people around the world. Recent advances in medical imaging based on deep learning can detect the disease earlier, up to 6 years before the onset of symptoms. They open a path to better treatments. Smart health and AI can predict Alzheimer.

A new system of artificial intelligence (AI) developed by US researchers could detect the first signs of Alzheimer’s six years before doctors can establish an effective diagnosis.

Scientists at the University of California, San Francisco trained their AI to detect imperceptible visual signs of metabolic changes in the brain to predict the development of Alzheimer’s disease.

Published in “Radiology”, the review of the North American Society of Radiology (RSNA), this study of 2,109 images of 1,002 patients already diagnosed found that the AI ​​had been able to discern with a precision of 100% the disease Alzheimer’s in images taken on average six years before diagnosis.

Alzheimer’s disease, the most common form of dementia, can contribute 60 to 70% of cases. There is currently no cure for Alzheimer’s disease, but studies have shown that deep learning can improve the ability of brain imaging to predict Alzheimer’s disease years before a real diagnosis, which would allow researchers find better ways to slow down, if not stop, the disease process.

50 MILLION PEOPLE WITH NEUROLOGICAL DISORDERS

When a radiologist reads an ultrasound, it is impossible to tell if a person will develop Alzheimer’s disease. Alzheimer’s disease is a problem that evolves differently from one patient to another, not all people with the same symptoms, and some cases may worsen more quickly than others. Doctors have no idea of ​​patients who will remain stable for some time or those who will quickly become unstable.

Research has linked changes in a person’s metabolism, such as increased glucose in some parts of the brain, to Alzheimer’s disease. Today, radiologists know how to identify specific biomarkers of the disease, but the metabolic changes are actually much more complex than one might think.

PREDICT ALZHEIMER ,A NEW APPROACH FOR BRAIN IMAGING

One of the most used diagnostic tools by doctors to identify the onset of Alzheimer’s disease is a type of brain imaging called PET 18-F-fluorodeoxyglucose. This analysis is traditionally used to identify several types of cancers, but in recent years has been shown to be useful in identifying Alzheimer’s disease, as well as many other types of dementia. When the algorithm was finally tested on a small independent set of brain scans, it was able to predict each case that progressed to Alzheimer’s disease on average about six years before the disease was finally diagnosed.

This new approach based on artificial intelligence uses graphics processors (GPU), particularly suitable for the execution of deep learning algorithms thanks to their parallel architecture. On this measure, the algorithm has clearly surpassed that of human radiologists. Currently, the use of PET is mainly limited to research studies and clinical trials, to ensure that new potential drugs are tested in the right people. PET is a powerful tool, but it is expensive and requires specialized facilities and expertise.

BETTER UNDERSTANDING THE EVOLUTION OF DISEASE

Whether imaging or clinical data, sensors or video games, the goal is to observe multiple patients over long periods to synthesize all this information in a digital model and dynamic aging brain. The goal is to better understand and, in the long term, predict how the disease develops; what are the anatomical but also metabolic changes in an affected brain.

 Artificial intelligence has the potential to help millions of people and is proving to be a useful tool for radiologists. Progress is remarkable. At present, there is no way to diagnose Alzheimer’s disease quickly and effectively. The use of such a system will allow physicians to detect the disease earlier, to better identify symptoms, to better understand behavior and progression, and to tailor treatments more precisely.