The evaluation and the anatomical-functional imaging of the heart represent one of the most important and complex fields in biomedical research. The heart, indeed, is the blood circulation’s motor and its proper work is a necessary condition for life. Many imaging techniques have been developed to study the heart, the cardiac magnetic resonance stands between them. The magnetic resonance technique allows, through the detection of the radio waves emitted by the hydrogen nucleous dipped in a precise magnetic field, to obtain informations concerning the soft tissues who arrange our body. This technique, who has really low invasiveness, during years has earned more and more space in the different clinical practiques. For the heart, particularly, the magnetic resonance allows to obtain precise and high resolution anatomycal informations of the cardiac walls. In the common practice it’s always segmented the cardiac wall who outlines the blood-wall interface with the resulting evaluation of many important indexes for the assessment and recognition of possible pathological endocardial movements like cardiac areas, the ejection fraction and the total volume. The manual segmentation exhib, however, the huge limit to be really laborious and user who execute it experience dependant and, consequently, suffers a strong inter-operator variability. So the biomedical research extended, during the last years, the research of techniques and algorithms who could allow the semi-automatic segmentation of the cardiac walls both in the trasversal section images called “short axis” and in the longitudinal ones called “long axis”. Today, the semi-automatic segmentation to segment the endocardium contours are a reality and produce better and better results. All, or almost all, the proposed algorithms don’t take in account the presence of the papillary muscles; omitting or including them in the cardiac wall recognition. The papillary muscles are musculuar appendices who project themselves in the cardiac cavitiy. Due to the position, form and size, most of the dedicated algorithms for their identification and consequent segmentation have a strong manual component asked to the user, mostly in the cases we’re they are segmented from the long axis projections. In the short axis sections, indeed, the papillary muscles exhibit a characteristic circular form repeated, concentrically, in the different projections. In the long axis projections, instead, the papillary muscle is cut longitudinally and exhibit a different form not only due to the characteristic anatomical properties of the patient but also due to the dependency form the particular phase of the cycle analysed. This thesis work propose a semi-automatic algorithm to detect and segment the papillary muscles from the long axis projections. The short axis will be used to recognize in what long axis sections the papillary muscle is visible. Subsequently, every pic is submitted to different masks built using the a priori anatomical knowledge of the papillary muscles and a previously executed endocardium segmentation. A thresholding algorithm, with thresholds obtained through a data classification with a clustering algorithm, is executed to obtain a preliminary segmentation. Subsequently, the segmentation is filtered applying morphologic operators, object’s area conditions, inter-projection of the obtained objects’ correlation conditions and a new classification to recognize the group membership of the obtained points. It has been decided to focus only on the left ventricle because, between them, it has more clinical valence. Cornerstone of this work has been the continuous research of the appropriate combination between a priori anatomical knowledge and the ability of the algorithm to adapt itself to the submitted data set. The final purpose is to arrive to a robust process capable to hit the target in different conditions taking in account the possible inter-patient differences.
La valutazione e l’imaging anatomo-funzionale del cuore rappresentano uno dei campi più importanti e complessi della ricerca biomedica. Il cuore, infatti, è il motore della circolazione sanguigna ed il suo corretto funzionamento è una condizione necessaria per la vita. Varie tecniche di imaging sono state sviluppate per lo studio del cuore, tra cui spicca la risonanza magnetica cardiaca. La tecnica di risonanza magnetica permette, tramite la detezione delle onde radio generate dai nuclei di Idrogeno immersi in un preciso campo magnetico, di ottenere informazioni riguardo i tessuti molli che compongono il nostro corpo. Questa tecnica, che gode di bassissima invasività, ha negli anni acquisito uno spazio sempre maggiore nelle diverse pratiche cliniche. Per il cuore, in particolare, la risonanza magnetica permette di ottenere informazioni anatomiche precise e ad alta risoluzione delle pareti cardiache. Nella pratica classica spesso si segmenta la parete endocardica che delinea l’interfaccia sangue-parete con la conseguente valutazione di molteplici indici importanti per la valutazione e il riconoscimento di possibili movimenti endocardici patologici come ad esempio il calcolo di aree cardiache, il volume di eiezione e il volume totale. La segmentazione manuale presenta, però, il grande limite di essere molto laboriosa e di dipendere dall’esperienza dell’operatore che la esegue e, di conseguenza, di soffrire di una forte variabilità inter-operatore. La ricerca biomedica si è quindi mossa, negli ultimi anni, nella ricerca di tecniche ed algoritmi che permettessero la segmentazione semi-automatica delle pareti endocardiche sia nelle immagini a sezione trasversale dette “assi corti” che in quelle a sezione longitudinali dette “assi lunghi”. Ad oggi, le tecniche per la segmentazione semi-automatica dell’endocardio sono una realtà e generano risultati sempre migliori. Tutti, o quasi, gli algoritmi proposti non tengono conto della presenza dei muscoli papillari; escludendoli o inglobandoli nel riconoscimento della parete endocardica. I muscoli papillari sono delle appendici muscolari che si proiettano nella cavità cardiaca. A causa della loro posizione e della loro forma e dimensione, la maggior parte degli algoritmi dedicati al loro riconoscimento e conseguente segmentazione ha una forte componente manuale richiesta all’utente, soprattutto nei casi in cui si decide di segmentare i papillari nelle proiezioni ad asse lungo. Nelle sezioni in asse corto, infatti, i muscoli papillari presentano una caratteristica forma circolare che si ripete, concentrica, nelle diverse proiezioni. Nelle proiezioni in asse lungo, invece, il muscolo papillare è tagliato longitudinalmente e presenta una forma differente non solo per le caratteristiche anatomiche proprie del paziente, ma anche per la dipendenza dalla fase del ciclo cardiaco presa in esame. Questo lavoro di tesi propone un algoritmo semi-automatico per la detezione e la segmentazione dei muscoli papillari nelle proiezioni in asse lungo. Verranno utilizzate le immagini in asse corto per il riconoscimento delle sezioni in asse lungo in cui il muscolo papillare è visibile. Successivamente, ogni immagine viene opportunamente sottoposta a diverse maschere costruite utilizzando le conoscenze anatomiche a priori dei muscoli papillari e una segmentazione dell’endocardio precedentemente eseguita. Un algoritmo di thresholding, con soglie ottenute tramite una classificazione dei dati tramite algoritmi di clustering, è eseguito per ottenere una segmentazione preliminare. Dopo, la segmentazione viene filtrata applicando operatori morfologici, condizioni sull’area degli oggetti, condizioni sulla correlazione inter-proiezione degli oggetti ottenuti ed una nuova classificazione per il riconoscimento del gruppo di appartenenza dei punti. Si è deciso di concentrarsi unicamente sul ventricolo sinistro poiché, tra i due, esso ha una valenza clinica maggiore. Punto cardine di questo lavoro è stata la continua ricerca dell’opportuno connubio tra conoscenze anatomiche a priori e capacità dell’algoritmo di adattarsi al set di dati di volta in volta sottoposto. Lo scopo finale è giungere ad un processo robusto e capace di centrare l’obiettivo in diverse condizioni e di tener conto delle differenze inter-paziente presenti.
Sviluppo di un algoritmo per la detezione dei muscoli papillari da immagini di risonanza magnetica
BLASI, PAOLO
2015/2016
Abstract
The evaluation and the anatomical-functional imaging of the heart represent one of the most important and complex fields in biomedical research. The heart, indeed, is the blood circulation’s motor and its proper work is a necessary condition for life. Many imaging techniques have been developed to study the heart, the cardiac magnetic resonance stands between them. The magnetic resonance technique allows, through the detection of the radio waves emitted by the hydrogen nucleous dipped in a precise magnetic field, to obtain informations concerning the soft tissues who arrange our body. This technique, who has really low invasiveness, during years has earned more and more space in the different clinical practiques. For the heart, particularly, the magnetic resonance allows to obtain precise and high resolution anatomycal informations of the cardiac walls. In the common practice it’s always segmented the cardiac wall who outlines the blood-wall interface with the resulting evaluation of many important indexes for the assessment and recognition of possible pathological endocardial movements like cardiac areas, the ejection fraction and the total volume. The manual segmentation exhib, however, the huge limit to be really laborious and user who execute it experience dependant and, consequently, suffers a strong inter-operator variability. So the biomedical research extended, during the last years, the research of techniques and algorithms who could allow the semi-automatic segmentation of the cardiac walls both in the trasversal section images called “short axis” and in the longitudinal ones called “long axis”. Today, the semi-automatic segmentation to segment the endocardium contours are a reality and produce better and better results. All, or almost all, the proposed algorithms don’t take in account the presence of the papillary muscles; omitting or including them in the cardiac wall recognition. The papillary muscles are musculuar appendices who project themselves in the cardiac cavitiy. Due to the position, form and size, most of the dedicated algorithms for their identification and consequent segmentation have a strong manual component asked to the user, mostly in the cases we’re they are segmented from the long axis projections. In the short axis sections, indeed, the papillary muscles exhibit a characteristic circular form repeated, concentrically, in the different projections. In the long axis projections, instead, the papillary muscle is cut longitudinally and exhibit a different form not only due to the characteristic anatomical properties of the patient but also due to the dependency form the particular phase of the cycle analysed. This thesis work propose a semi-automatic algorithm to detect and segment the papillary muscles from the long axis projections. The short axis will be used to recognize in what long axis sections the papillary muscle is visible. Subsequently, every pic is submitted to different masks built using the a priori anatomical knowledge of the papillary muscles and a previously executed endocardium segmentation. A thresholding algorithm, with thresholds obtained through a data classification with a clustering algorithm, is executed to obtain a preliminary segmentation. Subsequently, the segmentation is filtered applying morphologic operators, object’s area conditions, inter-projection of the obtained objects’ correlation conditions and a new classification to recognize the group membership of the obtained points. It has been decided to focus only on the left ventricle because, between them, it has more clinical valence. Cornerstone of this work has been the continuous research of the appropriate combination between a priori anatomical knowledge and the ability of the algorithm to adapt itself to the submitted data set. The final purpose is to arrive to a robust process capable to hit the target in different conditions taking in account the possible inter-patient differences.File | Dimensione | Formato | |
---|---|---|---|
07-2016-Blasi-Paolo.pdf
solo utenti autorizzati dal 14/07/2017
Descrizione: Testo della tesi
Dimensione
8.38 MB
Formato
Adobe PDF
|
8.38 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/123643