Detecting simulated skew deviation using a smartphone eye-tracking application (EyePhone): a feasibility study.

INTRODUCTION

Strokes presenting with acute dizziness (i.e., vestibular strokes) are four times more likely to be misdiagnosed compared with strokes presenting with typical symptoms such as weakness. The HINTS eye exam (Head Impulse test, Nystagmus, and Test of Skew) can improve the diagnostic accuracy for vestibular strokes with 98% Sensitivity and 97% specificity. As part of the HINTS battery, the test of skew evaluates the vertical misalignment of the eyes of a vestibular cause. Skew deviations of >5 diopters are highly specific for stroke. However, as experts who can accurately detect skew deviation are not readily available in the ED, there is a need for accessible methods to quantify skew deviation. We sought to investigate the feasibility of detecting and quantifying simulated skew deviations using our smartphone eye-tracking application (EyePhone).

METHODS

We enrolled healthy volunteers and recorded their eye movements using EyePhone. To simulate skew deviation, we used two targets on a wall 1 m away. As the examiner performed the cross-cover test, participants shifted their gaze between the upper and lower target whenever the occluder moved between their eyes, mimicking the refixation saccades seen in skew deviation. We began with targets separated by 20 cm at 1 m (representing 20 prism diopters of skew) and gradually reduced this distance until the targets overlapped, indicating the absence of skew. We used videos recorded by EyePhone using the front camera of an iPhone placed within 40 cm of the eyes. We developed in-house code using features extracted with Google's Mediapipe to analyze the eye movements after automatically detected eye uncoverings.

RESULTS

The correlation between the EyePhone measured and the designed misalignment suggests that EyePhone may be capable of detecting skew deviations of ≥5 diopters in a controlled experimental setting. Analysis showed a Spearman correlation of 0.93 (CI: 0.90-0.95) between the EyePhone measurements and the design-induced skew deviations. The calculated area under the receiver operator curve for discriminating skews≥5 diopters from smaller ones was 0.90.

DISCUSSION/

CONCLUSION

These findings suggest that EyePhone may potentially serve as a practical tool for the detection of clinically significant skew deviation following further validation in clinical groups.