Feature Extraction and Classification of Medication-Induced Hyperkinesia During Treatment of Parkinson's Disease
(2020) In Master's Theses in Mathematical Sciences FMSM01 20201Mathematical Statistics
- Abstract
- The recent progress in wearable sensor technology, signal processing and machine learning enables novel applications in fields such as automatic disease symptom tracking and classification. In this project, data from an ongoing Parkinson’s disease study at the Lund University Hospital is gathered and analyzed, and classification models of varying complexity for symptom severity are evaluated. The accelerom- eter and gyroscope data is provided by a single mobile phone fastened around each patient’s trunk and is complemented by a certified specialist’s score labels, in clin- ical dyskinesia rating scale (CDRS). Out of four available tasks performed by the patient during measurement, the two tasks “walking” and “sitting and describing a... (More)
- The recent progress in wearable sensor technology, signal processing and machine learning enables novel applications in fields such as automatic disease symptom tracking and classification. In this project, data from an ongoing Parkinson’s disease study at the Lund University Hospital is gathered and analyzed, and classification models of varying complexity for symptom severity are evaluated. The accelerom- eter and gyroscope data is provided by a single mobile phone fastened around each patient’s trunk and is complemented by a certified specialist’s score labels, in clin- ical dyskinesia rating scale (CDRS). Out of four available tasks performed by the patient during measurement, the two tasks “walking” and “sitting and describing a picture” were cut out and analyzed in-depth.
The most practically usable simulation scenario was found to be forming in- dividual models for the patients, with the best performing model proving to be an unsupervised feature extracting autoencoder combined with a linear discriminant. We could with descent accuracy distinguish between whole test signals in a binary- class scenario for a majority of patients and perform skillfully in the multi-class scenario, although not well enough for practical usability. The total patient-average, mean macro f1-score and accuracy obtained for binary-class classification of short (2 second) signal segments from the “describe picture”-task were 0.74 and 0.81 re- spectively. The corresponding mean macro f1-score and accuracy for the multi-class classification case were 0.52 and 0.65 respectively. (Less) - Popular Abstract (Swedish)
- Framsteg inom forskningsområden som maskininlärning, signalbehandling och sensorteknik möjliggör automatisk klassificering av sjukdomssymptom. Läkare kan bedöma symptom vid Parkinsons sjukdom genom att granska videomaterial. Kan vi lära datorer att göra samma bedömning endast utifrån data från sensorer?
Parkinsons sjukdom, som årligen drabbar ungefär 2000 personer i Sverige, leder till att nervceller som tillverkar signalsubstansen dopamin långsamt dör. Orsaken till att sjukdomen utbryter är fortfarande inte klarlagd och som medicin ges dopaminersättning, som lindrar symptom som stelhet och darrningar.
Medicinen är svår att dosera och kan i förlängningen trötta ut nervsystemet ytterligare. Därför är det av intresse att i framtiden... (More) - Framsteg inom forskningsområden som maskininlärning, signalbehandling och sensorteknik möjliggör automatisk klassificering av sjukdomssymptom. Läkare kan bedöma symptom vid Parkinsons sjukdom genom att granska videomaterial. Kan vi lära datorer att göra samma bedömning endast utifrån data från sensorer?
Parkinsons sjukdom, som årligen drabbar ungefär 2000 personer i Sverige, leder till att nervceller som tillverkar signalsubstansen dopamin långsamt dör. Orsaken till att sjukdomen utbryter är fortfarande inte klarlagd och som medicin ges dopaminersättning, som lindrar symptom som stelhet och darrningar.
Medicinen är svår att dosera och kan i förlängningen trötta ut nervsystemet ytterligare. Därför är det av intresse att i framtiden kunna skräddarsy medicindoser på ett noggrannare sätt. Denna process skulle underlättas om vi kunde göra fortlöpande mätningar på patienten och låta till exempel mobiltelefonen eller en smart klocka göra en bedömning av symptomen.
I detta examensarbete har vi skapat modeller som skattar symptomen som uppkommer till exempel då en patient får en hög dos medicin. Vi har använt sensordata samlat vid forskningsenheten för neurologi i Lund, tillsammans med en experts bedömningar. Vi har testat vanliga faltningsnätverk (CNN) och autoencoders tillsammans med linjära diskriminanter (AE-LDA), som båda är metoder baserade på artificiella neurala nätverk. Dessa metoder jämförs med enklare modeller såsom naiva bayesianska klassificerare (NBC). Innan datan färdas genom nätverken så beräknas bland annat skattningar av signalernas spektrum och signalernas energi, information som sedan används i modellerna.
Resultaten visar att metoden med AE-LDA fungerar bäst och att bästa resultat uppnås om varje patient får sin egen, unika modell. Metoden behöver utvecklas och förbättras för att bli tillräckligt träffsäker i en tillämpning. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9029701
- author
- Liljeroth, Erik LU
- supervisor
- organization
- course
- FMSM01 20201
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Parkinson's disease, neural network, autoencoder, classification, feature extraction
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMS-3399-2020
- ISSN
- 1404-6342
- other publication id
- 2020:E76
- language
- English
- id
- 9029701
- date added to LUP
- 2020-10-05 12:58:06
- date last changed
- 2021-06-08 17:11:32
@misc{9029701,
abstract = {{The recent progress in wearable sensor technology, signal processing and machine learning enables novel applications in fields such as automatic disease symptom tracking and classification. In this project, data from an ongoing Parkinson’s disease study at the Lund University Hospital is gathered and analyzed, and classification models of varying complexity for symptom severity are evaluated. The accelerom- eter and gyroscope data is provided by a single mobile phone fastened around each patient’s trunk and is complemented by a certified specialist’s score labels, in clin- ical dyskinesia rating scale (CDRS). Out of four available tasks performed by the patient during measurement, the two tasks “walking” and “sitting and describing a picture” were cut out and analyzed in-depth.
The most practically usable simulation scenario was found to be forming in- dividual models for the patients, with the best performing model proving to be an unsupervised feature extracting autoencoder combined with a linear discriminant. We could with descent accuracy distinguish between whole test signals in a binary- class scenario for a majority of patients and perform skillfully in the multi-class scenario, although not well enough for practical usability. The total patient-average, mean macro f1-score and accuracy obtained for binary-class classification of short (2 second) signal segments from the “describe picture”-task were 0.74 and 0.81 re- spectively. The corresponding mean macro f1-score and accuracy for the multi-class classification case were 0.52 and 0.65 respectively.}},
author = {{Liljeroth, Erik}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master's Theses in Mathematical Sciences}},
title = {{Feature Extraction and Classification of Medication-Induced Hyperkinesia During Treatment of Parkinson's Disease}},
year = {{2020}},
}