| Abstract (Eng): | Migraine is a primary headache disorder that affects over 1 billion people across the world, afflicting 18\% of women and 6\% of men. The costs of migraine in Europe alone are estimated to be between €50 billion and €111 billion in 2011. Migraine attacks have four different phases: the premonitory phase, aura phase, headache phase and postdrome phase. Some migraine sufferers can predict an upcoming migraine attack up to 12 hours before its start based on premonitory symptoms, i.e., non-headache symptoms. This master's dissertation aims to find digital biomarkers to detect the early stages of a migraine attack by searching for patterns in physiological data collected by a wearable. This study proposes an algorithm to find relevant motifs using matrix profile techniques. It defines these relevant motifs as patterns that occur frequently before headache attacks, but infrequently in time windows without headache attacks. Using this method, no relevant motifs were found, however, the search was not exhaustive, as it was only tested for one participant, in a limited search interval, and on one signal. To test this method for finding relevant motifs, a more graspable use case was proposed. The method for finding relevant motifs before migraine attacks was slightly adjusted to find relevant motifs during walking. The results of this adjusted method suggest that it is possible to find relevant motifs, but hyperparameters need to tuned correctly. |
|---|