Blog: Clinical Grade Eye Tracking: Precision, Accuracy and Robustness

Accuracy vs Precision

For several EYE-SYNC assessments, including Saccades, VOR, and VORx, eye-target synchronization is measured by reporting on the metrics of precision and accuracy. While these concepts are closely related, there are important distinguishing factors to master for any clinician utilizing these metrics in their workflow. At the highest level, the precision of eye movements can be thought of as the tightness of the cluster. Accuracy can be conceived of as the closeness of the cluster to the target.

Accordingly, eye movements that fall around a target may be any combination of accurate and precise. For example, eye movements that fall exactly on the target and are closely grouped would be both accurate and precise. Eye movements that fall around the target but in a loose manner would be accurate but imprecise. Eye movement that are in a tight cluster but far from the target would be precise but inaccurate. And finally, eye movements that are neither tightly clustered nor near the target would be both inaccurate and imprecise.

Precise and accurate eye movements can inform a clinician in a variety of ways. Importantly, consistently precise but inaccurate eye movements would indicate to the clinician that there is a systematic error in the person’s eye movements. Additionally, a series of inaccurate and imprecise eye movement would be indicative of a more general issue since the subject is making different errors every time.

When looking at eye tracking systems, the device performance is reported in degrees of variability for the device’s ability to capture accurately and precisely. Furthermore, the overall robustness of an eye tracking system is a synthesis of the precision and accuracy of the eye tracker across a varied population with different phenotypes and demographics. The most robust eye trackers will be precise even when there is significant variability in facial and eye features; including eye color, eye shape, eyelid shape, eyelid sag, eyelash length, and so on. Eye trackers can be confounded by inherent features such as very droopy eyelids in geriatric populations, but may also become imprecise or inaccurate due to individual artificial features such as contact lenses, mascara, and eye liner. Mascara in particular can confound eye trackers that rely on pupil measurements as the darkness is confused for a pupil. Other eye tracking methodologies, such as corneal reflection measurements, are more liable to be more noisy and corrupted by the presence of hard contact lenses.

For clinical grade eye tracking, it is essential that a system is able to record eye movements both precisely and accurately. The ability to capture data in such high fidelity must be applicable across a wide ranging population such that the measurements are broadly accessible rather than limited to a narrow subset. The combination of precision and accuracy, as well as methodology and additional measures such as data quality control, ensures robustness and gives clinicians access to meaningful data that can drive real clinical insights.