| Abstrakt: |
Objective: Parkinson's disease (PD), which is clinically characterized by motor symptoms, is a neurodegenerative disease accompanied by cognitive disorders. Resting state functional connectivity can be measured by functional magnetic resonance imaging (fMRI) method. The basic elements of functional connectivity (FC) are the nodes, which represent neuroanatomical areas, and the edges, which represent the functional relationship between them. Machine learning (ML) algorithms that learn complex relationship patterns in FC matrices obtained in this way are used to support the clinical diagnosis of PD. In this study, the relationship between the motor and cognitive symptoms of Parkinson's patients and the edge data of the patients' FCs was examined with the help of ML algorithms. Methods: Resting state fMRI (rs-fMRI) data of 55 Parkinson's patients (age: 60.2±8.5) followed in the Behavioral Neurology and Movement Disorders Unit of Istanbul Faculty of Medicine and Stroop Test, Judgement of Line Orientation Test (JLO), and Unified Parkinson's Disease Rating Scale (UPDRS) scores were evaluated. FC values between seed regions were calculated using CONN (https://web.conn-toolbox.org/) software. For the ML application, support vector machine (SVM) and random forest algorithm were used. Feature selection is included in all algorithms. Results: When the data set with FC edge values, Stroop test, JLO and UPDRS scores was evaluated with the SVM algorithm, various FC patterns predicted test and scale scores with 0.83, 0.90 and 0.63 estimation accuracy, respectively, and in the trial with the random forest algorithm, 0.83, 0.55 and 0.90 estimation accuracy has been obtained. Conclusion: The results reveal that neuropsychological test and scale scores can be predicted with high performance by evaluating various FC patterns with ML algorithms in PD. By expanding these analyzes, it may be possible to differentiate the neural basis of impairment in specific cognitive domains. [ABSTRACT FROM AUTHOR] |
