Diagnosis and prognosis of cardiovascular diseases by means of texture analysis in magnetic resonance imaging
NAGIOS: RODERIC FUNCIONANDO

Diagnosis and prognosis of cardiovascular diseases by means of texture analysis in magnetic resonance imaging

DSpace Repository

Diagnosis and prognosis of cardiovascular diseases by means of texture analysis in magnetic resonance imaging

Show simple item record

dc.contributor.advisor Bodí Peris, Vicente
dc.contributor.advisor Moratal Pérez, David
dc.contributor.author Larroza Santacruz, Andrés Martín
dc.contributor.other Departament d'Enginyeria Electrònica es_ES
dc.date.accessioned 2017-09-07T12:53:36Z
dc.date.available 2017-09-08T04:45:05Z
dc.date.issued 2017 es_ES
dc.date.submitted 11-09-2017 es_ES
dc.identifier.uri http://hdl.handle.net/10550/60645
dc.description.abstract Cardiovascular diseases constitute the leading global cause of morbidity and mortality. Magnetic resonance imaging (MRI) has become the gold standard technique for the assessment of patients with myocardial infarction. However, limitations still exist thus new alternatives are open to investigation. Texture analysis is a technique that aims to quantify the texture of the images that are not always perceptible by the human eye. It has been successfully applied in medical imaging but applications to cardiac MRI (CMR) are still scarce. Therefore, the purpose of this thesis was to apply texture analysis in conventional CMR images for the assessment of patients with myocardial infarction, as an alternative to current methods. Three applications of texture analysis and machine learning techniques were studied: i) Detection of infarcted myocardium in late gadolinium enhancement (LGE) CMR. Segmentation of the infarcted myocardium is routinely performed using image intensity thresholds. The inclusion of texture features to aid the segmentation was analyzed obtaining overall good results. The method was developed using 10 LGE CMR datasets and tested on a separate dataset comprising 5 cases that were acquired with a completely different scanner than that used for training. Therefore, this preliminary study showed the transferability of texture analysis which is important for clinical applicability. ii) Differentiation of acute and chronic myocardial infarction using LGE CMR and standard pre-contrast cine CMR. In this study, two different feature selection techniques and six different machine learning classifiers were studied and compared. The best classification was achieved using a polynomial SVM obtaining an overall AUC of 0.87 ± 0.06 in LGE CMR. Interestingly, results on cine CMR in which infarctions are visually imperceptible in most cases were also good (AUC = 0.83 ± 0.08). iii) Detection of infarcted non-viable segments in cine CMR. This study was motivated by the findings of the previous one. It demonstrated that texture analysis can be used to distinguish non-viable, viable and remote segments using standard pre-contrast cine CMR solely. This was the most relevant contribution of this thesis as it can be used as hypothesis for future work aiming to accurately delineate the infarcted myocardium as a gadolinium-free alternative that will have potential advantages. The three proposed applications were successfully performed obtaining promising results. In conclusion, texture analysis can be successfully applied to conventional CMR images and provides a potential quantitative alternative to existing methods. es_ES
dc.format.extent 166 p. es_ES
dc.language.iso en es_ES
dc.subject cardiac magnetic resonance es_ES
dc.subject myocardial infarction es_ES
dc.subject machine learning es_ES
dc.subject texture analysis es_ES
dc.title Diagnosis and prognosis of cardiovascular diseases by means of texture analysis in magnetic resonance imaging es_ES
dc.type info:eu-repo/semantics/doctoralThesis es_ES
dc.subject.unesco resonancia magnética es_ES
dc.subject.unesco análisis de datos es_ES
dc.subject.unesco tratamiento digital de imágenes es_ES
dc.subject.unesco diagnóstico por imagen es_ES
dc.embargo.terms 0 days es_ES

View       (6.890Mb)

This item appears in the following Collection(s)

Show simple item record

Search DSpace

Advanced Search

Browse

Statistics