Toward Intelligent Head Impulse Test: A Goggle-Free Approach Using a Monocular Infrared Camera.

Area of Science:

• Vestibular neuroscience
• Medical imaging analysis
• Artificial intelligence in healthcare

Background:

• The video head impulse test (vHIT) is the gold standard for assessing vestibular function by evaluating the vestibulo-ocular reflex (VOR).
• Traditional vHIT requires specialized, calibrated head-mounted goggles, limiting its accessibility and ease of use.
• There is a need for more automated, user-friendly methods for vestibular function assessment.

Purpose of the Study:

• To introduce an intelligent head impulse test (iHIT) system utilizing a monocular infrared camera.
• To develop and validate a deep learning-based video classification approach for vestibular function determination using iHIT.
• To offer a cost-effective and automated alternative to conventional vHIT.

Main Methods:

• An iHIT framework was established using a monocular infrared camera to capture patient test videos.
• A dataset, DiHIT, of head impulse test (HIT) video clips was created.
• A two-stage, multi-modal deep learning network was developed to analyze eye and head motion from facial keypoints for semicircular canal (SCC) identification and VOR abnormality determination (SCC qualitation).

Main Results:

• The deep learning model achieved 100% accuracy in predicting the specific semicircular canal being tested (SCC identification).
• Predictive accuracy for VOR abnormality (SCC qualitation) was 84.1% for horizontal SCCs and 79.0% for vertical SCCs.
• The iHIT system demonstrated high performance in classifying vestibular function.

Conclusions:

• The iHIT system, leveraging a monocular camera and deep learning, successfully assesses vestibular function.
• iHIT eliminates the need for specialized goggles and equipment calibration, enabling complete automation.
• This approach offers significant benefits including low cost and ease of operation, enhancing accessibility for users.