Pharmacopsychiatry 2011; 21 - A2
DOI: 10.1055/s-0031-1292443

Robust and stable automatic detection of rapid-eye movements in REM sleep

M Adamczyk 1, S Fulda 1, M Pawlowski 1, A Steiger 1, F Holsboer 1, E Friess 1
  • 1Max Planck Institute of Psychiatry, Munich, Germany

Sleep characteristics are candidates for predictive biomarkers in depressed patients. In particular, REM-sleep disinhibition is prominent in major depression and increased amounts of rapid eye movements (REMs) were demonstrated in these patients but also in individuals at high familial risk to develop the disease. Therefore, elevated REM density is a candidate for an endophenotype of depressive disorders. Efficient and reliable measurement is crucial for assessment of REMs in large samples. Both inter- and intra-scorer variability influences the results considerably. We therefore developed and validated an algorithm for REM detection. A large training set was used for method development and thresholds settings. The algorithm was then validated on a test set of 12 sleep recordings. The test set was scored independently by two experienced sleep scorers. Mean correlation between the experts was 0.91 (epoch-wise pearson's correlation averaged across nights). Comparison of automatic scoring with each of the scorer revealed mean correlation of 0.90 and 0.94, respectively. Correlation between scorers and computer algorithm for a subset of epochs where both experts agreed were highest (r = 0.96). Correlation for automatic scoring and all comparisons were clearly within the range of or superior to visual scoring. This allows for automatic detection of REMs with good performance and avoiding scorer biases.