Characteristic oscillatory brain networks for predicting patients with chronic migraine

• Published in: Journal of headache and pain Vol. 24; no. 1; p. 139
• Main Authors: Hsiao, Fu-Jung, Chen, Wei-Ta, Wu, Yu-Te, Pan, Li-Ling Hope, Wang, Yen-Feng, Chen, Shih-Pin, Lai, Kuan-Lin, Coppola, Gianluca, Wang, Shuu-Jiun
• Format: Journal Article
• Language: English
• Published: Milan Springer Milan 18.10.2023; Springer Nature B.V; BMC
• Subjects: Accuracy; Chronic headaches and medication overuse; Chronic migraine; Classification; Default mode network; Fibromyalgia; Headache; Internal Medicine; Learning algorithms; Machine learning; Magnetoencephalography; Medicine; Medicine & Public Health; Migraine; Neural networks; Neurology; Pain; Pain Medicine; Pain-related network; Resting-state oscillatory connectivity; Sensorimotor system; Chronic migraine; Pain-related network; Default mode network; Magnetoencephalography; Resting-state oscillatory connectivity; Machine learning
• ISSN: 1129-2377; 1129-2369; 1129-2377
• DOI: 10.1186/s10194-023-01677-z

To determine specific resting-state network patterns underlying alterations in chronic migraine, we employed oscillatory connectivity and machine learning techniques to distinguish patients with chronic migraine from healthy controls and patients with other pain disorders. This cross-sectional study included 350 participants (70 healthy controls, 100 patients with chronic migraine, 40 patients with chronic migraine with comorbid fibromyalgia, 35 patients with fibromyalgia, 30 patients with chronic tension-type headache, and 75 patients with episodic migraine). We collected resting-state magnetoencephalographic data for analysis. Source-based oscillatory connectivity within each network, including the pain-related network, default mode network, sensorimotor network, visual network, and insula to default mode network, was examined to determine intrinsic connectivity across a frequency range of 1–40 Hz. Features were extracted to establish and validate classification models constructed using machine learning algorithms. The findings indicated that oscillatory connectivity revealed brain network abnormalities in patients with chronic migraine compared with healthy controls, and that oscillatory connectivity exhibited distinct patterns between various pain disorders. After the incorporation of network features, the best classification model demonstrated excellent performance in distinguishing patients with chronic migraine from healthy controls, achieving high accuracy on both training and testing datasets (accuracy > 92.6% and area under the curve > 0.93). Moreover, in validation tests, classification models exhibited high accuracy in discriminating patients with chronic migraine from all other groups of patients (accuracy > 75.7% and area under the curve > 0.8). In conclusion, oscillatory synchrony within the pain-related network and default mode network corresponded to altered neurophysiological processes in patients with chronic migraine. Thus, these networks can serve as pivotal signatures in the model for identifying patients with chronic migraine, providing reliable and generalisable results. This approach may facilitate the objective and individualised diagnosis of migraine.