Multi-centre comparison of five eye movement detection algorithms

Area of Science:

• Sleep Medicine
• Neuroscience
• Biomedical Engineering

Background:

• Objective comparison of automatic eye movement detection algorithms is difficult due to varied recording setups.
• Previous studies lacked standardized data for evaluating different algorithms.

Purpose of the Study:

• To objectively compare the performance of five automatic eye movement detection algorithms.
• To assess algorithm performance across different electrode montages.
• To evaluate the impact of electroencephalography (EEG) delta activity cross-talk subtraction.

Main Methods:

• Five distinct eye movement detection algorithms were applied to a single, standardized dataset.
• Algorithm outputs were compared against visually scored data.
• True and false detection rates were analyzed at various threshold levels for both rapid and slow eye movements.
• Algorithm performance was tested using both their original and alternative electrode montages.
• The effect of subtracting EEG delta activity cross-talk was investigated.

Main Results:

• Algorithm performance varied, with best results achieved when using the electrode montage for which they were originally designed.
• Performance significantly decreased when algorithms were applied to non-native electrode montages.
• Subtracting EEG delta activity cross-talk demonstrably improved the accuracy of eye movement detections.
• True and false detection percentages were quantified for different algorithms and threshold settings.

Conclusions:

• Algorithm performance is highly dependent on the electrode montage used.
• Standardized data and methodology are crucial for reliable algorithm comparison.
• EEG delta activity cross-talk subtraction is a valuable technique for enhancing the accuracy of automatic eye movement detection during sleep.
• Further research should focus on developing montage-independent algorithms or robust cross-talk correction methods.