Challenging functional connectivity data: machine learning application on essential tremor recognition

• Published in: Neurological sciences Vol. 44; no. 1; pp. 199 - 207
• Main Authors: Saccà, Valeria, Novellino, Fabiana, Salsone, Maria, Abou Jaoude, Maurice, Quattrone, Andrea, Chiriaco, Carmelina, Madrigal, José L. M., Quattrone, Aldo
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 2023; Springer Nature B.V
• Subjects: Attention; Biomarkers; Brain; Brain mapping; Brain research; Cerebellum; Cerebellum - diagnostic imaging; Classification; Essential Tremor - pathology; Functional magnetic resonance imaging; Humans; Learning algorithms; Machine Learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Medicine; Medicine & Public Health; Nervous system; Neural networks; Neurology; Neuroradiology; Neurosciences; Neurosurgery; Original Article; Pathophysiology; Physiology; Psychiatry; Recognition, Psychology; Support vector machines; Time series; Tremor; Tremor (Muscular contraction); Visual perception; Resting-state fMRI; Support vector machine; Essential tremor; Machine learning
• ISSN: 1590-1874; 1590-3478
• DOI: 10.1007/s10072-022-06400-5

  Background and aims This paper aimed to investigate the usefulness of applying machine learning on resting-state fMRI connectivity data to recognize the pattern of functional changes in essential tremor (ET), a disease characterized by slight brain abnormalities, often difficult to detect using univariate analysis. Methods We trained a support vector machine with a radial kernel on the mean signals extracted by 14 brain networks obtained from resting-state fMRI scans of 18 ET and 19 healthy control (CTRL) subjects. Classification performance between pathological and control subjects was evaluated using a tenfold cross-validation. Recursive feature elimination was performed to rank the importance of the extracted features. Moreover, univariate analysis using Mann–Whitney U test was also performed. Results The machine learning algorithm achieved an AUC of 0.75, with four networks (language, primary visual, cerebellum, and attention), which have an essential role in ET pathophysiology, being selected as the most important features for classification. By contrast, the univariate analysis was not able to find significant results among these two conditions. Conclusion The machine learning approach identifies the changes in functional connectivity of ET patients, representing a promising instrument to discriminate specific pathological conditions and find novel functional biomarkers in resting-state fMRI studies.