Investigating the Use of Machine Learning to Detect Unilateral Arm Weakness
(2020) In Master's Theses in Mathematical Sciences FMAM05 20201Mathematics (Faculty of Engineering)
- Abstract
- Stroke is one of the leading causes of death and the most common cause for disability among the adult population in Sweden, due in part to the fact that many patients do not arrive at the hospital in time for treatment. A system that could detect unilateral arm weakness based on arm movement, a common symptom of stroke, is therefore desirable. In recent years there has been rapid development in the area of machine learning classification. This thesis set out to investigate if it was possible to construct machine learning models that, when trained on arm movement data, could classify unilateral arm weakness with sufficient performance. A total of 9 different supervised machine learning setups, including deep neural networks and classical... (More)
- Stroke is one of the leading causes of death and the most common cause for disability among the adult population in Sweden, due in part to the fact that many patients do not arrive at the hospital in time for treatment. A system that could detect unilateral arm weakness based on arm movement, a common symptom of stroke, is therefore desirable. In recent years there has been rapid development in the area of machine learning classification. This thesis set out to investigate if it was possible to construct machine learning models that, when trained on arm movement data, could classify unilateral arm weakness with sufficient performance. A total of 9 different supervised machine learning setups, including deep neural networks and classical models, were evaluated. To evaluate the models, two metrics were defined, the percentage of cases detected within 90 minutes (PCDT) and the false alarm rate per week (FAR per week). These metrics were estimated for all trained models when applied to a test set. All models were able to produce adequate performance given the right hyperparameters showing that it is possible to utilize machine learning to classify unilateral weakness. The fully convolutional network gave the best result as it obtained a PCDT of 79% and a FAR per week of only 2.4 on the test set. Among the evaluated models, the deep neural networks showed the most promise, but simpler designs obtained surprisingly good performance. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9014976
- author
- Persson, Emma LU and Lyckegård Finn, Eric LU
- supervisor
-
- Karl Åström LU
- organization
- course
- FMAM05 20201
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- machine learning, deep learning, InceptionTime, time series classification, multivariate time series
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMA-3408-2020
- ISSN
- 1404-6342
- other publication id
- 2020:E27
- language
- English
- id
- 9014976
- date added to LUP
- 2020-06-24 13:38:14
- date last changed
- 2020-06-24 13:38:14
@misc{9014976, abstract = {{Stroke is one of the leading causes of death and the most common cause for disability among the adult population in Sweden, due in part to the fact that many patients do not arrive at the hospital in time for treatment. A system that could detect unilateral arm weakness based on arm movement, a common symptom of stroke, is therefore desirable. In recent years there has been rapid development in the area of machine learning classification. This thesis set out to investigate if it was possible to construct machine learning models that, when trained on arm movement data, could classify unilateral arm weakness with sufficient performance. A total of 9 different supervised machine learning setups, including deep neural networks and classical models, were evaluated. To evaluate the models, two metrics were defined, the percentage of cases detected within 90 minutes (PCDT) and the false alarm rate per week (FAR per week). These metrics were estimated for all trained models when applied to a test set. All models were able to produce adequate performance given the right hyperparameters showing that it is possible to utilize machine learning to classify unilateral weakness. The fully convolutional network gave the best result as it obtained a PCDT of 79% and a FAR per week of only 2.4 on the test set. Among the evaluated models, the deep neural networks showed the most promise, but simpler designs obtained surprisingly good performance.}}, author = {{Persson, Emma and Lyckegård Finn, Eric}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Investigating the Use of Machine Learning to Detect Unilateral Arm Weakness}}, year = {{2020}}, }