Neurai-VN, A Real-world Multimodal Digital Phenotyping Dataset for Depression and Anxiety Disorders

ABSTRACT: Digital phenotyping (DP), integrating wearable and smartphone-based sensing, enables continuous and objective assessment of mental health in real-world settings. Despite its potential, DP datasets targeting mental health remain scarce in low- and middle-income countries (LMICs), particularly those combining clinician-validated diagnostic labels with structured self-report measures, limiting the generalisability, clinical validity, and equitable translation of DP models in these contexts. To address this gap, we present Neurai-VN, a real-world, high-resolution, multimodal dataset comprising passive sensing from wearable and smartphone devices, collected from 100 Vietnamese adults (aged 18–50) from the general population over two weeks. Participants were clinically screened and categorized into four mutually exclusive groups: individuals with major depressive disorder, individuals with generalized or social anxiety disorder, healthy controls, and individuals with other psychiatric conditions. The dataset integrates continuous wearable physiological signals, smartphone-derived behavioral data, clinician-assigned DSM-5 diagnostic and severity labels, and responses to validated self-report measures, including the PHQ-9, GAD-7, and brief daily mood assessments. To capture these data in real-world conditions, we deployed an in-house mobile application on participants’ personal iOS and Android devices. From the raw recording, we retrieved and implemented a robust pipeline to process sensor data into a standardized format. To this end, we derived and released 1,259 day-level records per participant across 14 sensing modalities, of which 8 were recorded at 1-minute resolution, and the remaining 6 were aggregated as daily summary measures, alongside 2,342 validated self-report entries. By providing richly annotated, multimodal data from an LMIC cohort, Neurai-VN dataset enables reproducible development and validation of AI models for depression and anxiety and facilitates discovery of digital biomarkers from real-world signals, addressing the scarcity of clinically validated mental health datasets in underrepresented populations. Latest Update: May, 15, 2026 Version Released Date Description Version 1.0.0 May, 15, 2026 Added P0XXX zip file, indicates individual data file.