Scientists from Tomsk State University (TSU) and Novosibirsk State University (NSU) have developed a new method for diagnosing early-stage depression based on graph neural networks (GNNs, designed for data processing). Its accuracy is 93%, according to the press service of the Russian Ministry of Education and Science.
For the first time in the world, researchers managed to combine electroencephalography (EEG) data and genetic markers in one algorithm. The neural network learned to detect the presence of the disorder.
Neda Firoz, an employee of the TSU Faculty of Physics, explained that a psychiatrist usually assesses a patient's condition using psychometric scales and surveys. Then, they prescribe an electroencephalogram (brain activity check) and a blood test for genetic abnormalities.
The use of graph neural networks allows these data to be integrated into a single system and to find correlations. This method will form the basis for automated depression diagnosis, the specialist added.
The computer notices complex combinations of signs that a psychiatrist simply cannot calculate without it.
The neural network provides a result based on pure mathematics. Therefore, the psychiatrist will receive a tool that excludes the human factor, Firoz concluded.
For the study, information on 383 patients was taken from the ICBrainDB dataset (prepared by TSU and NSU scientists). 34 of them had previously been diagnosed with depression. In the vast majority of cases, the neural network correctly diagnosed the condition.
Комментарии