Method, system and device for classifying and predicting functional magnetic resonance images

Abstract

The invention belongs to the technical field of computational medicines, particularly relates to a method, system and device for classifying and predicting functional magnetic resonance images, and aims at solving the problem of functional magnetic resonance image classification including time sequence information. The method comprises the steps of acquiring a functional magnetic resonance image group of a measured object which comprise a plurality of the functional magnetic resonance images; respectively performing feature dimensionality reduction on each brain region of each functional magnetic resonance image, and based on each of time sequences of the functional magnetic resonance images in the functional magnetic resonance image group, constructing brain region-time sequence feature matrix; classifying and predicting the brain region-time sequence feature matrix by a pre-trained classification model. The method achieves the functional magnetic resonance image classification including the time sequence information quickly and conveniently by computer means.

Description

technical field [0001] The invention belongs to the technical field of computational medicine, and in particular relates to a method, system and device for classifying and predicting functional magnetic resonance images. Background technique [0002] With the development of economy, health and medical level, the average life expectancy of people all over the world has been generally extended. At the same time, the burden of mental disorders continues to increase in all countries of the world, with significant health impacts and significant social, human rights and economic consequences. Clinically, no stable biomarkers have been found that can be used to assess the severity and cognitive level of mental illness. Therefore, the current diagnostic and classification standards for mental illness are based on clinical symptoms and behavioral descriptions, which have strong subjectivity. Moreover, some mental illnesses often show overlapping and overlapping clinical characteris...

AI Technical Summary

Problems solved by technology

[0004] The currently widely used time series feature extraction methods for fMRI images are mainly divided into two categories: the first is the feature extraction method based on the predefined brain template; the second is the data-driven feature extraction method; the template-based The feature extraction method is generally based on a fixed, predefined template (such as Automated AnatomicalLabelling (AAL), Brainnetome Atlas) to divide the brain area, and then extract the time series of each brain area. The advantage of this method is that it has good stability. , easy to migrate between different data, the disadvantage is that there is often a deviation between the predefined brain area and the real data; data-driven feature extraction methods (independent component analysis, etc.) can directly extract effective feature brain from the data area, this method can generally find more targeted functional networks than those based on brain templates. The disadvantage is that it needs to manually select features, and it is not easy to transfer learning between different data and

After extracting the time series, previous studies often calculate the correlation of the time series of different brain regions, and then construct the functional connection network, and use it as the input feature of the classification model. Such methods lose the original signal to a large extent. The time series information implicit in

With the rapid development of deep learning technology, the performance of deep learning models (cyclic neural networks, etc.) in analyzing time series (such as speech signals, natural language processing) has significantly exceeded traditional time series models (such as hidden Markov models, etc.) However, there is no research on the time series analysis of fMRI using deep learning at home and abroad.

In addition, the black-box nature of deep learning makes it difficult to effectively interpret its classification and diagnosis results, thus largely hindering its clinical application.

Images