Wearable-Derived Behavioral and Physiological Biomarkers for Classifying Unipolar and Bipolar Depression Severity

TL;DR

This study addresses the challenge of distinguishing unipolar from bipolar depression using wearable-derived biomarkers. It introduces the CALYPSO dataset, extracts multimodal features from HRV, EDA, accelerometer, and temperature signals, and benchmarks seven classifiers with leave-one-out cross-validation for subtype discrimination. Results show that accelerometer-based physical activity features, when paired with an MLP, achieve the highest accuracy (up to 96.77%), with temperature features also showing strong discriminative power (~93%), while HRV-based features are comparatively weaker due to signal noise. The findings demonstrate the potential of passive, continuous wearable monitoring to improve diagnostic precision and inform personalized treatment, though multimodal fusion requires careful feature engineering and advanced integration strategies.

Abstract

Depression is a complex mental disorder characterized by a diverse range of observable and measurable indicators that go beyond traditional subjective assessments. Recent research has increasingly focused on objective, passive, and continuous monitoring using wearable devices to gain more precise insights into the physiological and behavioral aspects of depression. However, most existing studies primarily distinguish between healthy and depressed individuals, adopting a binary classification that fails to capture the heterogeneity of depressive disorders. In this study, we leverage wearable devices to predict depression subtypes-specifically unipolar and bipolar depression-aiming to identify distinctive biomarkers that could enhance diagnostic precision and support personalized treatment strategies. To this end, we introduce the CALYPSO dataset, designed for non-invasive detection of depression subtypes and symptomatology through physiological and behavioral signals, including blood volume pulse, electrodermal activity, body temperature, and three-axis acceleration. Additionally, we establish a benchmark on the dataset using well-known features and standard machine learning methods. Preliminary results indicate that features related to physical activity, extracted from accelerometer data, are the most effective in distinguishing between unipolar and bipolar depression, achieving an accuracy of $96.77\%$. Temperature-based features also showed high discriminative power, reaching an accuracy of $93.55\%$. These findings highlight the potential of physiological and behavioral monitoring for improving the classification of depressive subtypes, paving the way for more tailored clinical interventions.

Figures (4)

• Figure 1: Overview of the proposed framework for classifying depression types based on physiological signals collected from the Empatica E4 wristband. The initial step involves pre-processing the raw sensor data to remove noise and artifacts, thereby enhancing the signal-to-noise ratio. Subsequently, time-domain, frequency-domain, and statistical features are extracted from the filtered signals. Finally, these extracted features are fed individually and combined to ML algorithms for the classification of unipolar and bipolar depression.
• Figure 2: Interview Room Setup for the Calypso Depression Dataset.
• Figure 3: A representative PSD for IBI signal showing The areas of VLF, LF, HF and VHF powers of the HRV.
• Figure 4: Visualization of an EDA signal. The top panel shows the raw and cleaned EDA signal. The middle panel displays the skin conductance response (SCR), highlighting the phasic component with identified SCR-onsets, SCR-peaks, and SCR-half recovery points. The bottom panel represents the skin conductance level (SCL), highlighting the tonic component of the EDA signal over time.