Primary Faculty Mentor Name

Ryan McGinnis

Project Collaborators

Ellen McGinnis, Jessica Hruschak, Nestor Lopez-Duran, Kate Fitzgerald, Katherine Rosenblum, Maria Muzik, Ryan McGinnis

Status

Undergraduate

Student College

College of Engineering and Mathematical Sciences

Program/Major

Biomedical Engineering

Primary Research Category

Engineering & Physical Sciences

Secondary Research Category

Health Sciences

Presentation Title

Playing with Bubbles can Identify Young Children with Internalizing Disorders

Time

1:00 PM

Location

Silver Maple Ballroom - Engineering & Physical Sciences

Abstract

Anxiety and depression are chronic conditions that affect relationships, development, and functioning, and can start as early as the preschool years. Current diagnostic assessments and referrals only capture the most severely impaired preschoolers, missing a large number who may develop additional clinical impairments. Herein, we consider a standardized task meant to elicit hedonic response by allowing children to interact with a rewarding stimulus; playing with bubbles. Wearable sensors are used to quantify child response to the task, and machine learning relates these objective behavioral measures to internalizing diagnosis.

Wearable sensor data were collected from an IMU secured around the head of each child while playing with bubbles for up to 3 minutes. IMU data were separated into two phases during the task: Initial (first 30 seconds) and Sustained (130-150 seconds). Children with internalizing disorders were identified via KSADS-PL with clinical consensus. Supervised learning was used to train classification models relating IMU-derived features to internalizing diagnosis based on data from N=48 children (N=14with an internalizing diagnosis). Performance of the classifiers was established using a leave-one-subject-out cross validation.

The model developed from data sampled during the Sustained phase of the bubbles task outperforms the model from the Initial phase (Accuracy: 0.85 vs. 0.42; Sensitivity: 1.00 vs. 0.86; Specificity: 0.79 vs. 0.24) and provides high sensitivity for identifying children with an internalizing diagnosis. Differences in sustained response to positive stimuli between individuals with and without internalizing disorders has been demonstrated previously in older subjects, providing support for these results.

These results suggest that wearable sensor data capturing a child’s motion for 150 seconds while playing with bubbles can be used to identify young children with an internalizing disorder with high sensitivity.

This document is currently not available here.

Share

COinS
 

Playing with Bubbles can Identify Young Children with Internalizing Disorders

Anxiety and depression are chronic conditions that affect relationships, development, and functioning, and can start as early as the preschool years. Current diagnostic assessments and referrals only capture the most severely impaired preschoolers, missing a large number who may develop additional clinical impairments. Herein, we consider a standardized task meant to elicit hedonic response by allowing children to interact with a rewarding stimulus; playing with bubbles. Wearable sensors are used to quantify child response to the task, and machine learning relates these objective behavioral measures to internalizing diagnosis.

Wearable sensor data were collected from an IMU secured around the head of each child while playing with bubbles for up to 3 minutes. IMU data were separated into two phases during the task: Initial (first 30 seconds) and Sustained (130-150 seconds). Children with internalizing disorders were identified via KSADS-PL with clinical consensus. Supervised learning was used to train classification models relating IMU-derived features to internalizing diagnosis based on data from N=48 children (N=14with an internalizing diagnosis). Performance of the classifiers was established using a leave-one-subject-out cross validation.

The model developed from data sampled during the Sustained phase of the bubbles task outperforms the model from the Initial phase (Accuracy: 0.85 vs. 0.42; Sensitivity: 1.00 vs. 0.86; Specificity: 0.79 vs. 0.24) and provides high sensitivity for identifying children with an internalizing diagnosis. Differences in sustained response to positive stimuli between individuals with and without internalizing disorders has been demonstrated previously in older subjects, providing support for these results.

These results suggest that wearable sensor data capturing a child’s motion for 150 seconds while playing with bubbles can be used to identify young children with an internalizing disorder with high sensitivity.