Evaluation of two algorithms for counting forward saccades in a reading task.

Area of Science:

• Ophthalmology
• Cognitive Science
• Biomedical Engineering

Background:

• Accurate analysis of eye movements, specifically forward saccades during reading, is crucial for understanding visual attention and cognitive processes.
• Existing algorithms for saccade detection rely on different metrics, such as velocity or jerk, with varying sensitivities to noise and artifacts.

Purpose of the Study:

• To compare the performance of two distinct algorithms for detecting forward saccades in eye movement recordings during a reading task.
• To evaluate the artifact rates and sources of error associated with velocity-based and jerk-based saccade detection methods.

Main Methods:

• Two algorithms were applied to eye movement data from 24 subjects reading German text on different displays.
• One algorithm analyzed eye movement velocity, while the second utilized the third derivative of eye position (jerk) for saccade detection.
• Data were collected at a mean reading speed of 258.5 words per minute.

Main Results:

• Both algorithms produced comparable artifact rates (2.5%) when detection thresholds were appropriately set and consistently applied.
• Artifact rates increased with threshold deviations from optimal settings for both velocity and jerk-based methods.
• The jerk-based algorithm exhibited greater inter-individual variation in artifact rates, with eye blinks identified as a significant artifact source.

Conclusions:

• Velocity and jerk-based saccade detection algorithms can achieve similar accuracy in reading tasks with proper threshold calibration.
• Eye blinks represent a major challenge for saccade detection, necessitating targeted artifact reduction strategies.
• Further research can focus on developing improved methods to mitigate blink-related artifacts in eye movement analysis.