Yanxia Kang ¹ Jianghao Zhao ² Yanli Zhao ³ Zilong Zhao ⁴ Yuan Dong ⁵ Manjie Zhang ⁶ Guimei Yin ^7* Shuping Tan ^3*

• ¹ Beijing HuiLongGuan Hospital, Beijing, China
• ² School of Computer Science and Technology, Taiyuan Normal University, Jinzhong, China
• ³ Center for Psychiatric Research, Beijing Huilongguan Hospital, Beijing, Beijing Municipality, China
• ⁴ School of Chemical Engineering and Technology, Sun Yat-sen University, Zhuhai Campus, Guangzhou, Guangdong Province, China
• ⁵ School of Computer Science and Technology, Taiyuan Normal University,, jinzhong, China
• ⁶ ,School of Computer Science and Technology, Taiyuan Normal University,, jinzhong, China
• ⁷ College of Computer Science and Technology, Taiyuan Normal University, Taiyuan, Shanxi, China

A multimodal persistent topological feature extraction and classification method is proposed to improve the recognition accuracy of first-episode schizophrenia patient. It can overcome the limitations of higher-order brain network analysis based on a single persistent feature (e.g., a persistent image) by exploiting the rich topological information generated during PH filtering. The experiment was conducted on resting-state EEG data from 198 subjects who met the study requirements at Huilongguan Hospital in Beijing. The subjects had a mean age of 30 years, a mean education of 14 years, and a sex ratio of 102 males to 96 females. Persistent topological features were extracted using adaptive thresholding during the PH filtrations. The distribution of these features was represented by heatmaps and persistence entropies, while the generation process of persistent features was interpreted using Betti curves and persistence landscapes. Finally, the classification performance of multimodal persistent topological features was evaluated using multiple machine learning classifiers. By selecting the classifier with the best performance, the multimodal persistent topological features were compared with traditional brain network features based on graph theory and single persistent topological features. The experimental results demonstrate that this method can provide high-dimensional, globally effective features of topological changes in first-episode schizophrenia patients, first-episode schizophrenia patient had significant changes throughout the persistent homology filtering compared to healthy subjects, the topological features extracted using the univariate feature selection algorithm achieved a classification accuracy of 94.6% using a combination of AC≥0.6 attributes. The proposed method has important clinical significance for the early identification and diagnosis of first-episode schizophrenia patients and offers a new research perspective for constructing higher-order functional connectivity networks and extracting topological structure features.