Work Description

Title: Deep Learning and statistical millisecond motor assessments of neurodevelopmental disorders Open Access Deposited

http://creativecommons.org/licenses/by-nc/4.0/
Attribute Value
Abstract
  • It is important to identify individuals with neurodevelopmental disorders (NDD) to facilitate early intervention but this goal is difficult to achieve with insufficient resources and providers with the necessary expertise. Autism spectrum disorder, attention-deficit/hyperactivity disorder, and their comorbidity are common NDD. Here we use high-definition kinematic sensors to study these NDD vs neurotypical (NT) conditions in individuals performing the reaching protocol. Using deep-learning (DL) techniques on the raw kinematic data, we differentiate these conditions with high accuracy. Analyzing the random motor fluctuations in the noise filtered data, we identify two quantitative NDD biometrics corresponding to clinical severity. Both DL and our statistical metrics show promise for initial screening and for estimation of NDD severity with high accuracy that may complement and enhance current clinical protocols.
Methodology
  • We studied the reaching movement paradigm of the cohort described below, using the XSENS MTw Awinda high-definition Bluetooth 120 Hz motion capture wireless sensors ( www.xsens.com/products/mtw-awinda) placed in a glove. The process was repeated an average of 100 times moving forward and 100 times backwards over about 15 minutes. 
Description
  • The data set consists of movement kinematic data from 92 participants that are neurotypical or have certain neurodevelopmental disorders (Autism, Attention Deficit/Hyperactivity Disorder, and their comorbidity). The data is of roughly 100 trials of a reaching paradigm per participant. The raw data consists of the linear acceleration (a), the roll (R), pitch (P), and yaw (Y), and the rate of change of the roll, pitch, and yaw. Description of the data and file structure. There are three primary folders: Data, Matlab Code, and the Deep Learning Code.

  • Data: Each folder corresponds to one participant. Inside is their motion data (sensor_data.csv) and a file describing their diagnosis and severity (diagnosis.txt). The columns of the csv are: a in x, a in y, a in z, dR/dt, dP/dt, dY/dt, R, P, Y The diagnosis.txt file contains two lines: the first is their diagnosis (ASD, ADHD, A^2, NT), the second is their severity (HF for high functioning, MF for mid functioning, LF for low functioning, and NA for not applicable).

  • Matlab Code: Information on how to analyze participants can be found in README.txt file within the folder. The Matlab code was written and run on MATLAB_R2023b. The signal processing toolbox is required.

  • Deep Learning Code: Information on how to perform the deep learning analysis can be found in README.txt file within the folder. The deep learning code is written in Python3 and relies on PyTorch.
Creator
Depositor
Contact information
Funding agency
  • National Science Foundation (NSF)
Keyword
Date coverage
  • 2019-09-13 to 2023-04-07
Citations to related material
  • K. Doctor, C. McKeever, A. Phadnis, D. Wu, J. Nurnberger Jr., M. Plawecki, & J.V. Jose. Deep learning and statistical millisecond motor assessments of neurodevelopmental disorders. Science, In Submission.
Publisher
Resource type
Last modified
  • 12/27/2023
Language
License
To Cite this Work:
Doctor, K., McKeever, C., Phadnis, A., Wu, D., Plawecki, M., Nurnberger Jr, J., Jose, J. Deep Learning and statistical millisecond motor assessments of neurodevelopmental disorders [Data set]. Indiana University - DataCORE.

Relationships

Files (Count: 2; Size: 360 MB)

Thumbnail Title Original Upload Last Modified File Size Access Actions
Data.zip 2023-12-27 2023-12-27 360 MB Open Access
Deep_Learning_Code.zip 2023-12-27 2023-12-27 Open Access

Download All Files

All files in this dataset will be included in the zip download.
Files can be downloaded individually in the "Files" panel above.

Best for data sets < 3 GB. Downloads all files plus metadata into a zip file.