Comparing In-ear EOG for Eye-Movement Estimation With Eye-Tracking: Accuracy, Calibration, and Speech Comprehension.
Level 3 - non-randomized controlled study
Controlled within-subject laboratory comparative study without reported randomization
PubMed 35844213 · doi:10.3389/fnins.2022.873201
What was done
Twenty-five hearing-impaired participants were fitted with custom ear molds containing electro-oculography (in-ear EOG) electrodes and wore eye-tracking glasses while viewing a video of a two-person dialogue (Diapix task) presented with directional sound and background noise at 60 dB SPL. Participants completed speech comprehension tests under three acoustic steering conditions: no steering, in-ear EOG steering, and conventional eye-tracking steering. Detected visual attention toward a speaker triggered a 6 dB amplification of that speaker. In-ear EOG channel selection (one electrode pair out of 36) was guided by a novel calibration procedure, and speaker estimation accuracy was compared against optical eye-tracking.
What was found
In-ear EOG estimated the attended speaker with a mean accuracy of 68% (range: 50% to 89%). Calibration metrics correlated significantly with data quality in offline simulation. Speech comprehension showed a statistically significant improvement of about 10% in both steering conditions compared to the no-steering condition, with no significant difference between in-ear EOG and conventional eye-tracking steering. Better comprehension under in-ear EOG steering was significantly correlated with higher estimation accuracy.
Why it matters
The findings demonstrate the feasibility of using in-ear electro-oculography to steer directional amplification in hearing assistive devices, matching the comprehension benefit of bulky eye-tracking systems in a controlled setting.
Limits
The sample size was small (n = 25 hearing-impaired participants). The experiment was conducted in an artificial laboratory setting with life-size video dialogue rather than real-world interactive conversations. Tracking accuracy varied widely across participants (50% to 89%) with a modest mean of 68%. The abstract does not evaluate long-term electrode stability, head motion artifacts, or device comfort.