Unipolar and Bipolar Depression Detection and Classification Based on Actigraphic Registration of Motor Activity...

Area of Science:

• Digital Health
• Computational Psychiatry
• Machine Learning in Healthcare

Background:

• Wearable sensors generate rich data beyond basic activity tracking, offering potential for objective health monitoring.
• Current depression assessment methods, like the Montgomery-Asberg Depression Rating Scale (MADRS), are subjective and labor-intensive.
• There is a need for objective, scalable methods to detect and monitor mental health conditions like depression.

Purpose of the Study:

• To introduce a novel dataset of sensor-derived motor activity from depressed patients and healthy controls.
• To investigate the efficacy of machine learning algorithms, including UMAP, for detecting depression from sensor data.
• To explore the relationship between motor activity patterns and depressive symptom severity.

Main Methods:

• Collected continuous sensor data (motor activity) from 32 healthy controls and 23 patients with unipolar/bipolar depression.
• Employed UMAP (Uniform Manifold Approximation and Projection) for unsupervised dimensionality reduction.
• Utilized various machine learning classifiers (e.g., nearest neighbors, SVM, Gaussian process, random forest, QDA) for depression detection.

Main Results:

• Initial model training showed challenges with class imbalance (Cohen Kappa < 0.1).
• A second experiment achieved high accuracy (0.991) with UMAP dimensionality reduction and machine learning.
• The combination of UMAP and neural networks yielded the best performance in depression detection.

Conclusions:

• Sensor data, particularly motor activity, holds significant potential for objective depression monitoring.
• UMAP combined with machine learning classifiers offers a promising approach for accurate depression detection.
• Further research correlating sensor-derived data with clinical depression ratings can enhance understanding and treatment.