MAD saccade: statistically robust saccade threshold estimation via the median absolute deviation.

Area of Science:

• Ophthalmology
• Neuroscience
• Computer Science

Background:

• Saccade detection is essential for analyzing eye-tracking data.
• Current velocity/acceleration threshold methods are sensitive to outliers, impacting accuracy.
• Existing methods can be influenced by the saccade signal itself.

Purpose of the Study:

• To introduce a robust method for saccade detection using median absolute deviation (MAD).
• To improve the accuracy and reliability of automatic saccade detection algorithms.
• To address the limitations of traditional threshold-based methods in gaze analysis.

Main Methods:

• Modified an existing saccade detection algorithm (Nyström et al.).
• Incorporated the median absolute deviation (MAD) as a robust dispersion estimator.
• Evaluated performance using both simulated and human eye-tracking data.

Main Results:

• The modified algorithm demonstrated significant improvements in saccade detection.
• Observed an increase in true positives and a decrease in false negatives.
• Performance gains were particularly notable under higher levels of data noise.

Conclusions:

• Robust estimators like MAD offer a superior alternative for saccade detection.
• The proposed method enhances the reliability of gaze data analysis.
• Robust estimators are easily implementable in other automatic gaze classification algorithms.