Abstract
There is no objective biological indicator for the diagnosis of schizophrenia. Machine learning is used to classify functional magnetic resonance imaging (fMRI) data, the aim of which is to effectively improve the reliability of diagnostics for schizophrenia. The following points are often considered: 1) Extracting effective features from fMRI data. 2) Choosing an appropriate machine learning method. 3) Improving classification accuracy. In this paper, we propose a weighted deep forest model, which includes a weighted class vector, and a prediction class vector. In our experiment, we extract functional connection (FC) features from fMRI data. Then, we use principal component analysis (PCA) to reduce the dimension of FC features. For datasets with unbalanced data, we use SMOTE to balance the data. Finally, the datasets with balanced data are fed into the weighted forest model. Compared with the classification results obtained by traditional classifiers, our classification accuracy is better. This method will provide greater possibilities for assisting doctors in diagnosing schizophrenia. This paper has significance for the study of schizophrenia by helping doctors diagnose the disease.
Highlights
All Functional magnetic resonance imaging (fMRI) data were preprocessed by using a toolbox named DPABI [17], which can be freely downloaded from the web site
Comparing the results in the two tables, we find that the revised gcForest is better than the traditional gcForest on all three datasets, and data balancing improves the performance of gcForest on fMRI data
Experimental results show that our method has superior performance to that of the traditional gcForest on fMRI data
Summary
Cheng et al [6] obtained functional networks from fMRI data, and they used a linear support vector machine (SVM) to classify patients with schizophrenia and members of the control group. In 2017, gcForest was proposed by Zhou and Feng [12]; it was a cascade forest model based on a deep neural network It obtained new features through multigranularity scanning. The research on schizophrenia based on fMRI data still faces the following problems: First, human understanding of brain structure and function is not deep enough [5]. Sakai and Yamada [14] summarized that in 21 schizophrenia classification studies over the past five years (2014-2018), the average sample size was 208, and the median sample size was 147 For such a small amount of data, choosing an appropriate machine learning method plays a key role. (4) We use SMOTE to balance the data in the training set before the training set is fed into the multi-grained scanning layer, which avoids the problem of unbalanced data
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
