Javascript must be enabled for the correct page display

Improving a neural network to classify movement disorders.

de Vries, Tanja (2020) Improving a neural network to classify movement disorders. Master's Internship Report, Computing Science.

[img]
Preview
Text
WMCS021-15_2020_Tanja_de_Vries.pdf

Download (2MB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (96kB)

Abstract

The aim of this project was to improve an existing Long Short Term Memory Recurrent Neural Network on movement disorders. The network was implemented by project Next Move in Movement Disorders, a collaboration between ZiuZ Visual Intelligence and the University Medical Centre Groningen. The network classifies patients based on 3D video data of them performing certain tasks. In this report we perform several experiments to get more clarity on the problematic areas of the network. The focus was mostly on myoclonus and tremor and partially on dystonia. We did an elaborate search on the possible feature vectors and values of the parameters. With these experiments we were not able to improve the classifier. After that, we explored the data itself. We found that the movement in the task is too dominant compared to the small involuntary movements that manifest due to the movement disorder. In the end we were able to find satisfactory results for a task that did not involve movement, here the network could distinguish myoclonus from dystonia. However, it could not distinguish myoclonus from tremor. More research will be needed to implement a network that can classify the disorders correct independent from the task and disorders we choose. The report is finished with an extensive discussion on future work.

Item Type: Thesis (Master's Internship Report)
Supervisor name: Biehl, M.
Degree programme: Computing Science
Thesis type: Master's Internship Report
Language: English
Date Deposited: 26 Aug 2020 12:27
Last Modified: 26 Aug 2020 12:27
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/23229

Actions (login required)

View Item View Item