Feature Extraction by QuickReduct Algorithm: Assessment of Migraineurs Neurovascular Pattern

Migraine is a complex disease with neurovascular implications. The accurate assessment of the migraineur's cerebrovascular status can be used to optimize and personalize therapy and ensure a better quality of life. Migraine has been associated to a disregulation in the carbon dioxide concentration, which can be measured by near-infrared spectroscopy (NIRS). In this paper we present a novel and automated strategy for feature extraction from NIRS time and frequency parameters. We built a large dataset by measuring 26 features for each of the 80 subjects we analyzed (51 migraineurs with aura, 14 without aura, and 15 healthy controls). We applied the QuickReduct algorithm (QRA) for the automated selection of the relevant features. Results were compared to the features extracted by the conventional ANOVA analysis. An artificial neural network (ANN) was used to perform classification in both cases (QRA and ANOVA). QRA returned 9 variables, ANOVA only 3. Only one feature was common to the two techniques: the increase of oxygen during a voluntary breath-holding. QRA coupled to ANN reached a classification accuracy of 97.5%, whereas ANOVA plus ANN gave about 75%. We plan to adopt this automated feature extraction strategy for the reduction of the dataset dimension in vascular characterization of migraineurs. The selection of the most important features find its importance in the analysis of complex systems and in monitoring protocols, where reduction ensures light computational burden and speeds up the time required for the patients assessment