Watch-Dog: Detecting Self-Harming Activities From Wrist Worn Accelerometers

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Clinical Psychology

Background:

• Suicide remains a significant concern in psychiatric facilities, with over 1800 inpatient suicides reported in the US in 2012.
• Current manual observation methods for preventing self-harm are resource-intensive and often insufficient.
• There is a growing need for effective, automated systems to monitor and detect self-harming activities in real-time within clinical settings.

Purpose of the Study:

• To introduce the Watch-Dog system, an innovative solution for detecting self-harming activities among psychiatric inpatients.
• To address the limitations of manual patient observation in preventing patient self-harm.
• To provide a practically deployable system for enhancing patient safety in clinical environments.

Main Methods:

• The Watch-Dog system utilizes tiny accelerometer sensors worn on patients' wrists to collect activity data.
• An efficient algorithm classifies user activity as either active or dormant.
• A novel decision selection algorithm, incorporating random forests and continuity indices, performs fine-grained activity classification.

Main Results:

• The Watch-Dog system demonstrated high classification accuracy in detecting self-harming activities.
• Achieved classification accuracies of [insert accuracy 1], [insert accuracy 2], and [insert accuracy 3] for same-user 10-fold cross-validation, cross-user 10-fold cross-validation, and cross-user leave-one-out evaluations, respectively.
• The system was evaluated using data from 11 subjects engaged in various activities, including self-harming behaviors.

Conclusions:

• The Watch-Dog system presents a practical and timely solution for detecting self-harming behaviors in psychiatric inpatients.
• The proposed system is deemed practically deployable, offering a significant advancement in patient safety monitoring.
• This research addresses a critical need for improved methods to prevent inpatient suicides in clinical settings.