The application of machine learning models based on nodal integrated topological attributes in the recognition of obsessive-compulsive disorder
10.3760/cma.j.cn371468-20240806-00356
- VernacularTitle:基于节点综合拓扑属性的机器学习模型在强迫症识别中的应用研究
- Author:
Shuaiqi ZHANG
1
;
Yangyang LIU
;
Pei LIU
;
Ningning DING
;
Zixuan LIU
;
Haisan ZHANG
Author Information
1. 河南省精神病医院（新乡医学院第二附属医院）磁共振科，新乡市多模态脑影像重点实验室，新乡市精神影像工程技术研究中心，新乡 453002
- Publication Type:Journal Article
- Keywords:
Obsessive-compulsive disorder;
Graph theory;
Data dimensionality reduction;
Machine learning;
Integrated node topology attributes
- From:
Chinese Journal of Behavioral Medicine and Brain Science
2025;34(5):426-432
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To create nodal integrated topological attributes (NITA) index and explore its application value in obsessive-compulsive disorder (OCD) identification by combining with machine learning model.Methods:Sixty-nine patients with OCD and 69 healthy volunteers matched with gender, age and years of education from the Second Affiliated Hospital of Xinxiang Medical University who met the enrollment criteria from January 2022 to September 2023 were included in the study.Their whole-brain functional magnetic resonance imaging (MRI) data were collected and preprocessed to construct the brain functional network, and the global and nodal topological attributes were extracted as the two sets of training features for the support vector machine (SVM), random forest and gradient boosting tree, and the better features were selected by comparing the classification results of the three machine learning models. The selected features were downgraded using principal component analysis algorithm, and the above models were trained again to filter out the models that were compatible with the new dimensional features. Finally, the new dimensional features with statistically significant differences in brain regions were screened and used to train the adapted model. SPSS 20.0 software was used to process relevant data, and independent sample t-test was used for inter group comparison. Results:Each machine learning model trained based on node topological attribute metrics was higher than the global metrics in terms of accuracy, recall, F1 value and AUC, and the average accuracy of the former was higher than that of the latter by about 10.00%. The node topology attribute metrics were downscaled and named NITA, which can synthesize about 95.00% of the feature information of node topology attribute metrics on average. SVM was finally chosen as the fitness model for NITA (accuracy of 86.00%, recall of 87.00%, F1 value of 0.86, AUC of 0.92). Compared with healthy controls, the differences in NITA in the medial superior frontal gyrus, middle frontal gyrus, ventral inferotemporal gyrus, caudal inferior parietal lobule, medial precuneus, insula hypergranular cellular area, caudal cuneus gyrus, inferior occipital gyrus, caudal hippocampus, dorsal caudate nucleus, and several subregions of the superior temporal gyrus and the thalamus were statistically significant in the OCD group (all P<0.05, FDR-corrected). Training the NITA of the above brain regions as features yielded the optimal model FDR-NITA-SVM, which had an accuracy of 91.38% in the training group and 90.00% in the test group. Conclusion:NITA can be used as a potential imaging marker for recognizing OCD.NITA abnormal brain regions are key nodes for information exchange and integration among brain networks in OCD patients.
