FIELD: medicine; data processing.
SUBSTANCE: invention relates to a method for graphed neural network classification for the absence or presence of major depressive disorder according to fMRI data. In the method, a functional MRI signal of the brain is recorded in a resting state in subjects of a control group and subjects with diagnosed major depressive disorder, as a result of which, in accordance with the automatic atlas of anatomical markup AAL3, a correlation matrix of size 166*166, each element of correlation matrix stores Pearson correlation coefficient, 20 threshold values are applied to the matrix elements in range from 0.0 to 0.95 with step of 0.05, wherein the correlation coefficient values below the threshold value are zeroed out and 20 sets of matrices of different sparsity degrees are obtained from the initial matrix, matrices are binarized by assigning value "1" for correlation coefficients other than zero, obtained 20 sets of binarized matrices are interpreted as graph adjacency matrices, for each set with a characteristic threshold value, constructing a distribution of relative frequencies based on the values of the shortest paths of the graph for both groups of subjects, for each distribution, calculating the value d* - length of shortest path with maximum value of t-statistics based on t-criterion for independent samples with correction for multiple comparison task, on 20 sets of binarized matrices of different degrees of sparsity, obtained at the previous stages of the method, constructing and training three graph neural networks with one output, which can be in states "0" and "1", and with number of convolutional graph layers equal to d*, d*-1 and d*+1, calculating the metric value F1 for each of the three constructed graph neural networks, and as a classifier selecting a network with a maximum of three value F1, supplying to the input of the classifier an adjacency matrix of the person being tested, for which it is necessary to conclude on the presence or absence of major depressive disorder: when a value of "0" is obtained at the output of the classifier, a conclusion is made on the absence of major depressive disorder in the person being tested, when a value of "1" is obtained at the output of the classifier, a conclusion is made on the presence of a major depressive disorder in the person being tested.
EFFECT: high accuracy of determining the presence of major depressive disorder by fMRI data.
1 cl, 5 dwg
