Biomarkers Selection based on FS-TNNR in Schizophrenia.

Schizophrenia (SZ) is a chronic mental illness that severely affects people's thoughts, feelings and behaviors. As time goes by, the symptoms will become more and more serious, clinically manifested as the confusion of thinking and speech, delusions, hallucinations, etc., which seriously endanger human health and life. Studies have shown that SZ is associated with abnormal brain regions and abnormal functional connections in brain regions. In recent years, magnetic resonance imaging (MRI) technology has been widely used in the medical field and scientific research, which contributes to the study of SZ. However, MRI data usually has a small sample size but a large number of features. How to effectively reduce the data dimension and accurately detect the biomarkers of SZ is of great significance in the diagnosis and further research of SZ. In this paper, truncated nuclear norm regularization (TNNR) is introduced into feature selection (FS) to form an embedded feature selection method. The non-convex low-rank advantage of the truncated nuclear norm as a regularization term is used to filter out those features that contribute the most to the identification data in SZ magnetic resonance data containing a large number of features and study the abnormal regions and abnormal functional connections in the SZ brain. The results show that the abnormal brain regions of SZ are mainly distributed in the supramarginal gyrus, cingulate gyrus, frontal gyrus, precuneus, frontal gyrus and caudate, and the abnormal functional connections are mainly related to the superior parietal gyrus, superior occipital gyrus, caudate nucleus, middle frontal gyrus, insula, supplementary motor area and precuneus. These abnormalities may lead to incorrect information processing in the brain, which provides further evidence for cognitive impairment in SZ.