Smartphone eye-tracking with deep learning: Data quality and field testing

Gancheng Zhu ¹, Zehao Huang ¹, Xiaoting Duan ¹

⁰Center for Psychological Sciences, Zhejiang University, 148 Tianmushan Rd., Hangzhou, 310028, China.

Behavior Research Methods +

|

June 18, 2025

View abstract on PubMed

Summary

Smartphone eye-tracking, powered by deep learning, offers comparable accuracy to gold-standard systems. This technology shows promise for detecting depressive symptoms with 76.67% accuracy in clinical applications.

Area Of Science

Computer Vision
Neuroscience
Mobile Health

Background

Eye-tracking is crucial for attention measurement across various fields.
Advancements in AI and mobile computing enable computer vision-based eye tracking on smartphones.

Purpose Of The Study

To present a real-time smartphone eye-tracking system using deep neural networks.
To benchmark its performance against a gold-standard eye tracker.
To evaluate its potential in clinical applications, specifically for depressive symptom assessment.

Main Methods

Developed a deep neural network trained on 7.4 million facial images for eye tracking.
Benchmarked the system against an EyeLink eye tracker with 32 participants.
Conducted a field test with 98 volunteers using visual tasks on a smartphone to assess depressive symptoms.

Main Results

Smartphone eye-tracking demonstrated comparable accuracy (1.32° vs. 1.20°) to the EyeLink tracker, though with lower precision (0.177° vs. 0.028°).
The system achieved 76.67% accuracy in predicting depressive symptoms based on visual task performance.

Conclusions

Smartphone eye-tracking provides quality data suitable for scientific and clinical use.
This technology has significant potential for accessible and widespread application in mental health assessment.

Keywords:

Computer vision Data quality Depressive symptoms Eye-tracking Smartphone