diplomsko delo
Marko Verdenik (Author), Franc Jager (Mentor)

Abstract

V diplomskem delu predstavimo avtomatično analizo elektromiograma maternice s koherenčno funkcijo, ki je ena izmed nelinearnih tehnik procesiranja signalov. Uporabili smo posnetke mednarodne referenčne baze TPEHG DB, ki vsebuje 300 posnetkov. Signale smo predprocesirali z devetimi različnimi Butterworthovimi pasovno-prepustnimi filtri in v izogib faznemu popačenju uporabili dvosmerno shemo filtriranja. Ločevanje skupin je potekalo v dveh variantah, ločevanje med zgodaj snemanimi in ločevanje med pozno snemanimi posnetki. Izračunali smo koherenčno funkcijo med vsemi pari posnetkov za vsako od variant. Računali smo jo med močnostnima spektroma signalov. Za oceno koherence za celotno frekvenčno območje smo izbrali dve cenilki - mediano amplitude in integral. Enosmerna analiza varianca ali ANOVA je pokazala, katere skupine posnetkov so primerne za ločevanje prezgodnjega in terminskega poroda. Frekvenčna območja in signale, katerih p-vrednosti so manjše od 0,05, smo uporabili za klasifikacijo posnetkov. Za klasifikacijo smo uporabili Bayesov klasifikator, odločitvena drevesa in klasifikator, ki smo ga empirično sestavili sami. Opazili smo, da se koherenca med terminskimi porodi med frekvenčnima območjema 1-2,5 Hz in 2,5-3,5 Hz znatno zmanjša, medtem ko se koherenca prezgodnjih porodov znatno ne spremeni. Ocenjevanje zmogljivosti klasifikacije je potekalo na tri načine - na učni množici, po principu učna-testna množica in s pristopom "izpusti enega". Najboljšo oceno klasifikacije smo dobili z uporabo odločitvenih dreves na učni množici, na frekvenčnem območju 0,3-2,5 Hz, kjer je bila občutljivost 95 %, specifičnost in natančnost pa 98 %. Malo slabše rezultate smo dobili z uporabo lastnega klasifikatorja. Občutljivost je bila med 58 % in 63 %, specifičnost pa med 61 % in 65 % za izbrane filtre in kanale. Klasifikacija z Baysovim klasifikatorjem pa ni pokazala vzpodbudnih rezulatov, z občutljivostjo blizu 0 %.

Keywords

prezgodnji porod;elektrohisterogram;koherenčna funkcija;ločevanje skupin;klasifikacija;klasifikator;računalništvo;računalništvo in informatika;računalništvo in matematika;interdisplinarni študij;univerzitetni študij;diplomske naloge;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [M. Verdenik]
UDC: 621.3:618.4(043.2)
COBISS: 1536277699 Link will open in a new window
Views: 1603
Downloads: 459
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Separating groups of uterine electromiographic records with term and pre-term delivery using coherence function
Secondary abstract: In this thesis we present automatic analysis of electromyogram of uterus (electrohysterogram) using coherence function which is one of non-linear signal processing techniques. We used records of international reference database TPEHG DB (Term-Preterm Electrohysterogram DataBase), which contains 300 electrohysterogram records. We preprocessed signals with nine different band-pass Butterworths filters with forward-backward filtering to avoid zero phase shift. Separation of groups took place in two variants, among early recorded and among late recorded records. We calculated coherence function between all pairs of records for each of variants. For calculation we used power spectrum of signals. Coherence estimation for whole frequency range, was made with two techniques - median amplitude and integral. Analysis of variance or ANOVA showed which frequency ranges and signals are useful for preterm - term records separation. For records classification we used frequency intervals and signals with p-value less than 0,05. Evaluation of classification was made on Bayes classifier, decision trees and our own built classifier. We developed it empirically, based on coherence decreasing among term records from frequency range 1-2,5 Hz to 2,5-3,5 Hz. Performance evaluation of classification is done in three ways - on training set, on the principle of training-testing set and with the approach "omitted one". Best results were shown with decision tree at frequency range 0,3-2,5 Hz, where sensitivity was 95 %, specificity and accuracy were 98 %. With our own developed classifier we reach sensitivity between 58 % and 63 % and specificity between 58 % and 63 %. Classification using Bayes classifier did not show good results, having sensitivity close to 0 %.
Secondary keywords: preterm delivery;electrohysterogram;coherence function;group separation;classification;classifier;computer science;computer and information science;computer science and mathematics;interdisciplinary science;diploma;
File type: application/pdf
Type (COBISS): Undergraduate thesis
Study programme: 1000425
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 56 str.
ID: 8740307