Vis enkel innførsel

dc.contributor.advisorBongo, Lars Ailo
dc.contributor.authorGrønnesby, Morten
dc.date.accessioned2017-07-31T09:19:55Z
dc.date.available2017-07-31T09:19:55Z
dc.date.issued2016-06-01
dc.description.abstractLungs sounds has been used as a diagnostic tool for centuries. The usefulness of listening to lung sounds, or pulmonary auscultation, as a definite diagnostic method has been diminished by advances in medical imaging such as chest X-Ray, but these advanced methods also bring a higher monetary and time cost. In addition, when the severity of pulmonary conditions changes, audible symptoms change immediately, while x-ray imaging does not show the same immediate change. The stethoscope is still used as a screening method and has great potential for use in continuous monitoring, as it is a simple, non-invasive, and low cost method. Therefore, lung sounds are still important today, and there is a need for better training tools, and automatic analysis methods that can be integrated with stethoscopes to advance the technology of one of the most common medical tools today. As a part of Tromsøundersøkelsen, researchers are now recording lung sounds to create a gold standard database of lung sounds and categorizing them based on occurrences of abnormal sounds. They are investigating the validity of pulmonary auscultation as a diagnostic method. Earlier approaches have achieved good results in this field, but have lacked a large dataset and gold standard to validate performance in a general setting of clinical data. We present our approach to automatically analyze lung sounds, and classify abnormal sounds found in audio files of recorded breathing. We employed signal processing and machine learning techniques and implemented an analysis pipeline to perform the classification. We achieved a cross-validated F1-score of 83.5\% using a Support Vector Machine performing classification on window excerpts containing \textit{Crackles} from recordings of breathing. We also did preliminary evaluation the classification for \textit{Wheezes}, and found a F1-score of 64.6\%. With our pipeline we have also implemented a GUI for a web application that we can deploy as a working prototype. We believe that with this approach we have created a basis for a core technology, that can be integrated with mobile platforms to serve as a home monitoring device, training tool or medical equipment.en_US
dc.descriptionPubliseringsvalg endret etter henvendelse fra student. [LL, 27.07.2017]en_US
dc.identifier.urihttps://hdl.handle.net/10037/11260
dc.language.isoengen_US
dc.publisherUiT Norges arktiske universiteten_US
dc.publisherUiT The Arctic University of Norwayen_US
dc.rights.accessRightsopenAccessen_US
dc.rights.holderCopyright 2016 The Author(s)
dc.subject.courseIDINF-3981
dc.subjectVDP::Technology: 500::Information and communication technology: 550::Computer technology: 551en_US
dc.subjectVDP::Teknologi: 500::Informasjons- og kommunikasjonsteknologi: 550::Datateknologi: 551en_US
dc.subjectVDP::Technology: 500::Medical technology: 620en_US
dc.subjectVDP::Teknologi: 500::Medisinsk teknologi: 620en_US
dc.titleAutomated Lung Sound Analysisen_US
dc.typeMaster thesisen_US
dc.typeMastergradsoppgaveen_US


Tilhørende fil(er)

Thumbnail
Thumbnail

Denne innførselen finnes i følgende samling(er)

Vis enkel innførsel