MASTER Assignment
AUTOMATIC DETECTION OF USER ERRORS IN SPIROMETRY DATA USING MACHINE LEARNING TECHNIQUES AND THE ANALYSIS OF THE EFFECT OF METAPHORS ON THE QUALITY OF SPIROMETRY MEASUREMENTS
Type: Master M-CS, M-ITECH
Period: Dec, 2019- Jul, 2020
Student : Heerlien, I.R. (Iris, Student M-CS,M-ITECH)
Date Final project: July 17, 2020
Supervisors:
Abstract:
This study is part of the project SpiroPlay, which goal is to improve the quality of home spirometry tests. This study focuses on three areas of this project. Firstly, an error detection algorithm based on machine learning techniques was designed. The results show that only the binary model, classifying if an attempt contains an error or not, performs well enough to be used in real life (recall: 0.864). The second area of focus is assessing the inter- and intra-rater agreement between professionals detecting errors in spirometry attempts. The negative to minimal inter-rater agreements (-0.123 to 0.380), and the moderate to strong intra-rater agreements (0.648 to 0.860) show that as professionals detect different errors in spirometry data, the rules on which the error detection is based are not strict enough. Therefore, before a generic error detection algorithm can be designed, the rules should be sharpened. The third focus area of this research is the evaluation of the difference in quality of spirometry attempts when coached by a professional versus by a metaphor. The FVC, FEV1, PEF values, and the number of errors were compared. No significant differences were found, implying metaphors are a good coaching manner for home spirometry measurements.