Analiza biomedicinskih signala prikupljenih nosivim uređajima metodama strojnog učenja

master's thesis

• Title (english): Machine learning-based analysis of biomedical signals acquired by wearable devices
• Author: Silvio Podhraški
• Mentor: Hrvoje Džapo (mentor)
• Committee member: Hrvoje Džapo (predsjednik povjerenstva)
• Committee member: Mario Cifrek (član povjerenstva)
• Committee member: Željka Lučev Vasić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-02-21, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U radu je opisan i implementiran proces korištenja metoda strojnog učenja prilikom analize biomedicinskih signala, s fokusom na fotopletizmografski signal (PPG). Navedeni proces obuhvaća predobradu biomedicinskih signala, ekstrakciju i odabir značajki signala, korištenje modela strojnog učenja te evaluaciju modela. Cilj rada je istražiti uspješnost klasifikacije ili grupiranja pojedinih epoha signala s obzirom na razinu šuma, za što se koriste metode strojnoga učenja. Predobrada biomedicinskih signala obuhvaća korištenje pojasnopropusnog Butterworth filtra za uklanjanje šuma, podjelu signala na prozore, te označavanje signala korištenjem klasifikatora baziranog na pravilima. Izlučuju se statističke i domenske značajke. Provodi se odabir značajki radi postizanja boljih rezultata klasifikacije i grupiranja. Za odabir značajki, tj. transformaciju značajki, korišten je mRMR algoritam i Lasso regularizacija, odnosno PCA. Korišteni su modeli nadziranog i nenadziranog strojnog učenja implementirani u scikit-learn biblioteci. Implementacija je izvedena u programskom jeziku Python. Provedena je usporedna analiza performansi različitih modela s obzirom na odabrane značajke. Uspoređuje se učinkovitost klasifikacije modela s obzirom na definirane mjere uspješnosti modela (F1 rezultat, AUC). Na kraju se zaključuje o primjenjivosti strojnog učenja kao jedne od mogućnosti analize biomedicinskih signala.

Abstract (english)

This paper describes and implements the process of using machine learning techniques for biomedical signal analysis, with an emphasis on PPG signals. The process includes biomedical signal preprocessing, feature extraction and selection, using machine learning models and model evaluation. The goal is to explore whether the machine learning algorithms are successful in the classification and grouping of the signal regarding the presented noise in the signal. The preprocessing step includes using the band-pass Butterworth filter, windowing, and labeling using a rule-based classifier. Statistical and domain features are extracted. Feature selection (mRMR, Lasso regularization) and feature transformation (PCA) are also used to improve the classification results. Supervised and unsupervised learning models are used, which are implemented in the scikit-learn library. The Python programming language is used for implementation. Based on selected features, the performance analysis of multiple models is carried out. The paper analyzes model performance regarding various evaluation metrics (F1 score, AUC). In the end, the paper concludes the usability of machine learning for biomedical signal analysis.

• Keywords: strojno učenje; klasifikacija vremenskih serija; biomedicinski signali; varijabilnost srčanog ritma; fotopletizmografija; Python; Pandas; scikit-learn
• Keywords (english): machine learning; time series classification; biomedical signals; heart rate variability; photoplethysmography; Python; Pandas; scikit-learn
• Language: croatian
• URN:NBN: urn:nbn:hr:168:052675
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-07-13 10:28:53