Facial Landmark Analysis for Detecting Visual Impairment in Mobile LogMAR Test

Maximilian Kapsecker , Elena Mille , Florian Schweizer

IEEE Journal of Biomedical and Health Informatics +

|

March 3, 2025

View abstract on PubMed

Summary

This study shows that analyzing facial features like cheek squinting during a mobile vision test can help detect vision loss. This technology offers a new way for continuous, unobtrusive eye health monitoring.

Area Of Science

Ophthalmology
Biomedical Engineering
Computer Science

Background

Visual impairment is a significant global health concern requiring timely intervention.
Smartphones can capture facial gestures, potentially indicating vision changes like squinting.
Current vision tests are often conducted in clinical settings, limiting continuous monitoring.

Purpose Of The Study

To develop and evaluate a mobile health application for digital vision testing using facial feature analysis.
To identify specific facial landmarks that correlate with reading performance during a vision test.
To explore the potential of smartphone-based facial tracking for unobtrusive vision assessment.

Main Methods

A mobile application was created to perform a digital logMAR test and record facial features simultaneously.
37 participants completed the mobile eye test.
Facial landmarks were analyzed using exploratory data analysis and receiver operating characteristic (ROC) curves to assess discriminative power.

Main Results

Facial regions including the nose, mouth, and cheeks showed significant changes with reduced reading performance.
Cheek squinting demonstrated high discriminative power, with an area under the curve (AUC) of 0.82.
Specific facial landmarks can distinguish between correct, partial, and incorrect letter readings.

Conclusions

Tracking facial features via smartphone offers a promising method for continuous and unobtrusive vision assessment.
Facial feature analysis integrated into everyday applications could facilitate widespread vision screening.
Further large-scale studies in non-standardized environments are warranted to validate these findings.