A Correlative Classification Study of Schizophrenic Patients with Results of Clinical Evaluation and Structural Magnetic Resonance Images

Abstract

Patients with schizophrenia suffer from symptoms such as hallucination and delusion. There are currently a number of publications that discuss the treatment, diagnosis, prognosis, and damage in schizophrenia. This study utilized joint independent component analysis to process the images of GMV and WMV and incorporated the Wisconsin card sorting test (WCST) and the positive and negative syndrome scale (PANSS) to examine the correlation of obtained brain characteristics. We also used PANSS score to classify schizophrenic patients into acute and subacute cases, to analyze the brain structure differences. Finally, we used brain structure images and the error rate of the WCST as eigenvalues in support vector machine learning and classification. The results of this study showed that the frontal and temporal lobes of a normal brain are more apparent than those of a schizophrenia brain. The highest level of classification recognition reached 91.575%, indicating that the WCST error rate and characteristic changes in brain structure volume can be used to effectively distinguish schizophrenia and normal brains. Similarly, this result confirmed that the WCST and brain structure volume are correlated with the differences between schizophrenia and normal participants.

Accuracy of diagnostic classification algorithms using cognitive-, electrophysiological-, and neuroanatomical data in antipsychotic-naïve schizophrenia patients

“…Generally, the studies, which have reported very high accuracies, have first applied a statistical test to pre-select variables, which discriminate between groups on the outcome measure for the specific dataset (e.g. Chu et al ., 2016; Santos-Mayo et al ., 2017). A recent SVM study using sMRI cortical thickness and surface data from 163 first-episode, antipsychotic-naïve patients (mean age 23.5 years) and matched controls (mean age 23.6 years) revealed a diagnostic accuracy of 81.8% and 85.0%, respectively, for thickness and surface.…”

“…Previous machine learning studies have generated encouraging diagnostic accuracies >85% (e.g. Shen et al ., 2014; Chu et al ., 2016; Santos-Mayo et al ., 2017; Xiao et al ., 2017) as well as prediction of the clinical outcome (Zarogianni et al ., 2017). However, most previous studies have been unimodal and performed in medicated and more chronic patient samples, in which the variation in data is greater than at first illness presentation.…”

Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis

“… b Eighty-six AI models 30 , 32 , 34 , 36 , 42 , 44 , 46 , 53 , 59 , 60 , 61 , 63 , 69 , 70 , 72 , 81 , 82 , 88 , 91 , 117 , 149 , 151 , 156 , 162 , 163 , 164 , 169 , 170 , 172 , 180 , 182 , 187 , 191 , 197 , 212 , 213 , 220 , 221 , 247 , 254 , 258 , 260 , 261 , 263 , 269 , 272 , 278 , 280 , 283 , 284 , 285 , 287 , 290 , 301 , 305 , 307 , 319 , …”

Machine Learning e Processos Diagnósticos com Avaliação Neuropsicológica: Uma Revisão Integrativa

“…No que tange ao uso de distintas modalidades de dados, deve-se ressaltar o papel fundamental das neuroimagens, as quais representaram a principal modalidade em articulação com dados de demais naturezas, o que foi observado em diversas pesquisas (Chu, Huang, Jian, Hsu, & Cheng, 2016;De Marco, Beltrachini, Biancardi, Frangi, & Venneri, 2017;Haller et al, 2013;Patel et al, 2015;Segovia et al, 2014). Destacase inclusive a possibilidade de tal articulação entre dados genéticos e avaliações cognitivas.…”