Evaluation of an accessible automatic sleep spindle detector

An important component of sleep medicine is assessing a patient’s polysomnogram, which includes sleep staging. The American Academy of Sleep Medicine (AASM) requires that sleep staging must take into account the number of sleep spindles, a short burst of neural oscillatory activity, demonstrating the need for good spindle detection. Currently this is done manually by sleep experts, but it is time consuming, motivating research into the creation of automatic spindle detectors. In this thesis, the sleep spindle detector known as “Yet Another Spindle Algorithm” (YASA) is evaluated, and compared to the spindle detector that it’s based on, known as A7. Furthermore, we put forward three hypotheses to put to the test, the first one being that YASA has similar performance to A7; the second being that YASA performs similarly when analyzing the frontal EEG channels (useful for self-applied somnography) compared to the central channels (the channels typically used for spindle detection); and the third hypothesis being that YASA performs better when analyzing N2 sleep data, compared to sleep data from other parts of the night. The results show that YASA performs similarly as A7, with F1-scores of 0.69 and 0.70, respectively. YASA performs similarly when analyzing frontal channels and central channels, with F1-scores of 0.61, and 0.62, respectively. However the recall is a bit higher when analyzing central channels compared to frontal channels (0.74 vs 0.64, respectively), and the precision is a bit higher for frontal compared to central (0.58 vs 0.53, respectively). As we hypothesized, YASA does perform better on N2 sleep data compared to other stages (F1: 0.69 vs 0.42, respectively).

Author dc:creator

• Hannes Kristján Hannesson 1995-

Contributors dc:contributor

• Háskólinn í Reykjavík

• Tölvunarfræði
• Meistaraprófsritgerðir
• Reiknirit
• Mælitæki
• Svefnrannsóknir
• Computer science
• Algorithms
• Measurement
• Sleep
• Research

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/1946/39415

OAI identifier oai:identifier

oai:skemman.is:1946/39415