Algorithm for early diagnosis of Alzheimer's disease

Artificial intelligence is able to detect changes in the brain several years before the onset of symptoms of Alzheimer's disease

Marina Astvatsaturyan, Echo of Moscow

A computer program using noninvasive magnetic resonance imaging data reveals changes in the connections between different parts of the brain, and this is possible at least ten years before the diagnosis of Alzheimer's disease is made - the main cause of senile dementia, leading to loss of memory and cognitive functions.

Different research groups in different countries are looking for methods for the earliest possible diagnosis of the disease. Although there is currently no cure for Alzheimer's disease, the drugs being created are more likely to be more effective the sooner the patient starts taking them. Early diagnosis will also allow you to start changing your lifestyle earlier in order to slow down the progress of the disease.

For early diagnosis of Alzheimer's disease Nicola Amoroso, Marianne La Rocca (Marianna La Rocca) and their colleagues from The University of Bari in Italy has developed a machine learning algorithm that recognizes structural changes in the brain caused by the disease.

To begin with, they trained the algorithm using 67 brain tomograms, 38 of them belonged to patients with Alzheimer's disease, and 29 belonged to healthy people. The source of the scan data was the database of a multicenter study called the Alzheimer's Disease Neuroimaging Initiative at the University of Southern California in Los Angeles.

The idea of the training was for the algorithm to master the exact classification and be able to distinguish a sick brain from a healthy one. Each tomogram was divided into small sections and the neural connections between them were analyzed without any assumptions about the ideal size of the sections for diagnosis. Thus, it was found that the most accurate classification of the disease by the algorithm is obtained when the volume of the compared brain regions is in the range from 2250 to 3200 cubic millimeters, which successfully coincided with the size of the anatomical structures associated with the disease – the amygdala and the hippocampus.

Then the scientists tested the algorithm on the second portion of tomograms from 148 people: 52 healthy, 48 patients with Alzheimer's disease and 48 with moderate cognitive impairment, who developed the disease later – in the range from 2.5 to nine years. The accuracy with which artificial intelligence distinguished a healthy brain from one affected by Alzheimer's disease was 86 percent.

The results of Italian scientists are presented on the arXiv preprint server (Brain structural connectivity atrophy in Alzheimer's disease).

Portal "Eternal youth" http://vechnayamolodost.ru  22.09.2017