Details

Autor(en) / Beteiligte

• D'Ostilio, Kevin
• Lisicki Martinez, Marco
• SCHOENEN, Jean
• MAGIS, Delphine

Titel

A MACHINE-LEARNING CLASSIFIER FOR EPISODIC MIGRAINE BASED ON VISUAL EVOKED GAMMA BAND ACTIVITY

Ort / Verlag

Blackwell Science

Erscheinungsjahr

2016

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Introduction: Objective and reliable biomarkers of migraine may be of interest for diagnosis and research purposes. Neuroimaging-based machine-learning classifiers are promising but hampered by availability and cost issues. Conversely, evoked potential are of easy access and affordable. They have provided increasing evidence that sensory information processing is impaired in migraine. We have used gamma band oscillations (GBOs) of visual evoked potentials (VEPs) to compute a machine-learning neural network classifier in episodic migraine. Materials and methods: We analyzed GBOs from VEPs (6x100 responses). Recordings were performed in two matched samples: a training sample composed of 43 migraine patients (EM) and 20 healthy volunteers (HV) and a validating sample of 18 EM and 10 HV. A logistic regression model of the training sample was performed to evaluate the relevance of the predictor variables. Ten neural networks were automatically generated based on late component frequency, n3-p4 and p4-n4 slopes, 1st block n1-p2 amplitude and age. Results: The logistic regression model of the training sample reached a significant classification rate of 79% (EM: 88%; HV: 60%, p¼0.002). The best neural network was able to classify the groups with an accuracy of 73% in the training phase and 89% in the subsequent validation (success rate HV: 90%; EM: 88%). The mean global accuracy within the training and validating samples were 69% (63–78%) and 84% (82–89%). Conclusions: This machine-learning neural network classifier based on visual GBOs provides an accurate and costefficient tool for objective migraine diagnosis. Further training and validation studies with new cohorts are warranted