PupilSense: Detection of Depressive Episodes Through Pupillary Response in the Wild
Rahul Islam, Sang Won Bae
TL;DR
Depression assessment remains challenging due to reliance on self-report and laboratory settings. This work introduces PupilSense, a mobile, deep-learning pipeline that passively captures pupillary responses via a smartphone’s front camera and estimates the pupil-iris ratio to identify depressive episodes in naturalistic contexts. In a proof-of-concept study with 25 participants and 528 labeled days, a Top Significant Features model achieves AUROC $0.71$, outperforming several sensor-only baselines and supporting the feasibility of unobtrusive, real-world depression monitoring using ubiquitous mobile devices. The findings demonstrate that pupillometry can complement existing digital phenotyping tools, offering a path toward continuous, privacy-conscious mental health care in everyday life.
Abstract
Early detection of depressive episodes is crucial in managing mental health disorders such as Major Depressive Disorder (MDD) and Bipolar Disorder. However, existing methods often necessitate active participation or are confined to clinical settings. Addressing this gap, we introduce PupilSense, a novel, deep learning-driven mobile system designed to discreetly track pupillary responses as users interact with their smartphones in their daily lives. This study presents a proof-of-concept exploration of PupilSense's capabilities, where we captured real-time pupillary data from users in naturalistic settings. Our findings indicate that PupilSense can effectively and passively monitor indicators of depressive episodes, offering a promising tool for continuous mental health assessment outside laboratory environments. This advancement heralds a significant step in leveraging ubiquitous mobile technology for proactive mental health care, potentially transforming how depressive episodes are detected and managed in everyday contexts.