|
Title:
|
AUDIO CONTEXT CLASSIFICATION FOR DETERMINING BLOOD PRESSURE SELF-MEASUREMENT ADHERENCE |
|
Author(s):
|
Stefan Wagner, Peter Ahrendt, Thomas S. Toftegaard, Olav W. Bertelsen |
|
ISBN:
|
978-972-8939-70-0 |
|
Editors:
|
Mário Macedo |
|
Year:
|
2012 |
|
Edition:
|
Single |
|
Keywords:
|
Blood pressure, self-measurement, machine learning, pervasive healthcare, data quality. |
|
Type:
|
Full Paper |
|
First Page:
|
105 |
|
Last Page:
|
114 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Blood pressure self-measurement (BPSM) is used in the diagnosis of hypertension and requires the patient to follow a range of recommendations in order to be considered valid for diagnostic use. One recommendation specifies that the patient must remain silent during measurements as talking may bias the measurement. Current state-of-the-art blood pressure (BP) devices cannot verify whether the patient complies. We suggest using audio context classification based on feature extraction and classification using an artificial neural network classifier to detect patients talking during measurement. For this purpose, we have developed an experimental algorithm and a software evaluation framework to obtain experimental data. We trained the algorithm using machine learning techniques with voice data from 80 unique test-subjects recorded in the laboratory setting, in order to classify talking and silence as two distinct modalities. Laboratory data from another 20 test-subjects were used to evaluate the algorithms efficiency. The algorithm was integrated into an integrated BPSM solution and evaluated with further 22 test-subjects. Results indicate that audio context classification is feasible with 99.0% correctly classified results in the laboratory and 97.3% in the integrated BPSM solution evaluation setup. |
|
|
|
|
|
|
