The cerebrovascular system is a central topic in neurological and neurosurgical research since it is the site of several pathologies which can lead to stroke, and it also affects the safety of minimally invasive interventions in neurosurgery, where vessels must be avoided to prevent the risk of internal haemorrhage. In this contest, efficient tools for subject-specific morphometric and hemodynamic analysis are needed to support clinical assistance in all its phases. In the present study, morphometric and hemodynamic information are combined to perform a subject-specific artery/vein (A/V) classification on a clinical dataset of segmented three-dimensional Digital Subtraction Angiographies (3D DSA) belonging to 60 patients with no cerebrovascular disorders which received Contrast Enhanced Cone Beam Computed Tomography (CE CBCT) for Stereo-ElectroEncephalography surgical planning. The analysis started with the extraction of the topological skeleton of each vascular mask, which is the core for vascular tree morphometric characterization. Skeletonization enabled vessels’ labelling and the estimation of their length and diameter. These steps were optimized and tested on a synthetic data-base and next applied to the clinical data-set. In particular, among four different solutions for diameter estimation, the best one was selected and applied to the clinical vascular masks. Moreover, the skeleton allowed the study of the connectivity of the tree to detect the direct connections and the pathways among the vessel branching sites (nodes). The hemodynamic information was extracted by further processing the CE CBCT acquisitions, obtaining a time-intensity curve for each voxel. Lastly, the fusion between the morphometric and hemodynamic information allowed the development of the new vessel-wise logic for A/V classification, by which entire vessels were classified according to the class of their skeleton segment obtained by applying a majority criterion. This method enabled the classification of 80% of the vascular voxels among the larger vessels (against the 60% of the previous voxel-wise approach on the same dataset) with values of sensitivity and specificity respectively of 92.6% and 93.1%, therefore the new approach outperformed the previous one without any loss of quality of the classification.
Il sistema cerebrovascolare è un tema centrale nella ricerca neurologica e neurochirurgica, in quanto è la sede di molte patologie che possono portare a eventi di ictus e inoltre influisce sulla sicurezza degli interventi di neurochirurgia mininvasiva, in cui è necessario evitare i vasi sanguigni per prevenire il rischio di emorragie. In questo contesto clinico, metodi automatici per la caratterizzazione morfometrica ed emodinamica specifica per il singolo paziente sono necessari per il supporto all’assistenza clinica in tutte le sue fasi. Lo studio propone la combinazione dell’informazione morfometrica ed emodinamica finalizzata alla classificazione arteria/vena specifica per un dataset clinico di angiografie sottrattive digitali (DSA) segmentate appartenenti a 60 pazienti provenienti da acquisizioni di Cone Beam Computed Tomography (CBCT), con somministrazione del mezzo di contrasto, per il planning operatorio di procedure di Stereo-Elettroencefalografia (SEEG). L’analisi prevede l’estrazione dello skeleton topologico di ogni maschera vascolare, che ha un ruolo principale nella caratterizzazione morfometrica, in quanto consente di isolare ogni vaso ed effettuare stime di lunghezza e diametro. Gli algoritmi relativi sono stati ottimizzati e testati su un dataset sintetico per poi essere applicati sul dataset clinico. In particolare, tra 4 diverse proposte, il miglior algoritmo per la stima del diametro dei vasi è stato applicato ad ogni maschera cerebrovascolare. Inoltre, lo skeleton ha consentito lo studio della connettività vascolare, al fine di individuare le connessioni dirette e traiettorie tra i vari punti di giunzione (nodi) dell’albero. L’informazione emodinamica è stata estratta dall’ulteriore elaborazione delle acquisizioni di CBCT, da cui sono state stimate curve tempo-intensità, (TIC) per ogni voxel. Infine, la fusione morfometrica ed emodinamica ha consentito lo sviluppo dell’approccio vessel-wise per la classificazione arteria/ vena, attraverso il quale interi vasi vengono classificati in base alla classe del loro skeleton relativo ottenuta secondo un criterio di maggioranza. Questo metodo ha fornito prestazioni molto migliori rispetto al precedente e senza alcuna perdita di qualità, in quanto ha garantito la classificazione dell’80% dei voxels vascolari (rispetto al precedente 60% sullo stesso dataset clinico) con valori di sensitività e specificità rispettivamente di 92.6% e 93.1%.
Morphometric and hemodynamic fusion for cerebrovascular artery/vein classification in DSA datasets
Marranchino, Vittorio
2022/2023
Abstract
The cerebrovascular system is a central topic in neurological and neurosurgical research since it is the site of several pathologies which can lead to stroke, and it also affects the safety of minimally invasive interventions in neurosurgery, where vessels must be avoided to prevent the risk of internal haemorrhage. In this contest, efficient tools for subject-specific morphometric and hemodynamic analysis are needed to support clinical assistance in all its phases. In the present study, morphometric and hemodynamic information are combined to perform a subject-specific artery/vein (A/V) classification on a clinical dataset of segmented three-dimensional Digital Subtraction Angiographies (3D DSA) belonging to 60 patients with no cerebrovascular disorders which received Contrast Enhanced Cone Beam Computed Tomography (CE CBCT) for Stereo-ElectroEncephalography surgical planning. The analysis started with the extraction of the topological skeleton of each vascular mask, which is the core for vascular tree morphometric characterization. Skeletonization enabled vessels’ labelling and the estimation of their length and diameter. These steps were optimized and tested on a synthetic data-base and next applied to the clinical data-set. In particular, among four different solutions for diameter estimation, the best one was selected and applied to the clinical vascular masks. Moreover, the skeleton allowed the study of the connectivity of the tree to detect the direct connections and the pathways among the vessel branching sites (nodes). The hemodynamic information was extracted by further processing the CE CBCT acquisitions, obtaining a time-intensity curve for each voxel. Lastly, the fusion between the morphometric and hemodynamic information allowed the development of the new vessel-wise logic for A/V classification, by which entire vessels were classified according to the class of their skeleton segment obtained by applying a majority criterion. This method enabled the classification of 80% of the vascular voxels among the larger vessels (against the 60% of the previous voxel-wise approach on the same dataset) with values of sensitivity and specificity respectively of 92.6% and 93.1%, therefore the new approach outperformed the previous one without any loss of quality of the classification.| File | Dimensione | Formato | |
|---|---|---|---|
|
2023_07_Marranchino_tesi+Executive_summary_01.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: TESI + EXECUTIVE SUMMARY
Dimensione
13.55 MB
Formato
Adobe PDF
|
13.55 MB | Adobe PDF | Visualizza/Apri |
|
2023_07_Marranchino_Executive_summary_02.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: EXECUTIVE SUMMARY
Dimensione
1.39 MB
Formato
Adobe PDF
|
1.39 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/10589/211387