The availability of realistic 3D models of the liver, with identified vessels and tumors, is a key element in planning liver surgery interventions. This study aims to develop an automatic medical image processing pipeline for segmentation of hepatic blood vessels and tumors, starting from 3D contrast-enhanced CT volumes. The proposed project involves the use of the 3D Slicer platform and the RVesselX module. For the automated identification of vascular structures, multiscale vascular filters including Frangi, Sato, Jerman, OOF and Meijering were applied followed by post-processing techniques to improve the morphological quality of segmentations. Liver tumors were segmented separately, allowing a complete reconstruction of the anatomical-pathological picture. Validation of the effectiveness of filter parameters was a central part of the work. A systematic approach based on Design of Experiments was used, with the aim of exploring the influence of key parameters (such as sigma, alpha, beta and gamma) on the effectiveness of the filters and on the robustness of the segmentation in the presence of qualitative variations of the CT scans. The parametric combinations generated through DOE were applied on a dataset of abdominal CT scans from the Medical Segmentation Decathlon, an open source collection containing abdominal images with segmentations of vessels and tumors. The segmentation results, obtained after the application of each parametric configuration, were compared with the corresponding masks of liver vessels using similarity metrics, such as the Dice Coefficient. The segmented models were subsequently exported in STL format for future holographic visualization, in order to provide advanced support to medical personnel for pre-operative planning.
La disponibilità di modelli tridimensionali realistici del fegato, con vasi e tumori identificati, è un elemento chiave nella pianificazione di interventi di chirurgia epatica. Questo studio si propone di sviluppare una pipeline di elaborazione automatica di immagini medicali per la segmentazione dei vasi sanguigni epatici e dei tumori, partendo da volumi 3D TAC con mezzo di contrasto. Il progetto proposto prevede l’utilizzo della piattaforma 3D Slicer e del modulo RVesselX. Per l’identificazione automatizzata delle strutture vascolari sono stati applicati filtri vascolari multiscala tra cui Frangi, Sato, Jerman, OOF e Meijering seguiti da tecniche di post-processing per migliorare la qualità morfologica delle segmentazioni. I tumori epatici sono stati segmentati separatamente, consentendo una ricostruzione completa del quadro anatomico-patologico. La validazione dell’efficacia dei parametri dei filtri ha rappresentato una parte centrale del lavoro. Si è utilizzato un approccio sistematico basato su Design of Experiments, con l’obiettivo di esplorare l’influenza dei parametri chiave (come sigma, alpha, beta e gamma) sull’efficacia dei filtri e sulla robustezza della segmentazione in presenza di variazioni qualitative delle TAC. Le combinazioni parametriche generate tramite DOE sono state applicate su un dataset di TAC addominali provenienti dal Medical Segmentation Decathlon, una raccolta open source contenente immagini addominali con segmentazioni di vasi e tumori. I risultati delle segmentazioni, ottenuti in seguito all’applicazione di ciascuna configurazione parametrica, sono stati confrontati con le corrispondenti maschere di vasi epatici mediante metriche di similarità, come il Coefficiente di Dice. I modelli segmentati sono stati successivamente esportati in formato STL per una futura visualizzazione olografica, al fine di fornire al personale medico un supporto avanzato alla pianificazione pre-operatoria.
Segmentazione Vascolare Automatizzata con Filtri Multi-Scala e Validazione Parametrica per la Chirurgia Epatica
PINTO, YLENIA
2024/2025
Abstract
The availability of realistic 3D models of the liver, with identified vessels and tumors, is a key element in planning liver surgery interventions. This study aims to develop an automatic medical image processing pipeline for segmentation of hepatic blood vessels and tumors, starting from 3D contrast-enhanced CT volumes. The proposed project involves the use of the 3D Slicer platform and the RVesselX module. For the automated identification of vascular structures, multiscale vascular filters including Frangi, Sato, Jerman, OOF and Meijering were applied followed by post-processing techniques to improve the morphological quality of segmentations. Liver tumors were segmented separately, allowing a complete reconstruction of the anatomical-pathological picture. Validation of the effectiveness of filter parameters was a central part of the work. A systematic approach based on Design of Experiments was used, with the aim of exploring the influence of key parameters (such as sigma, alpha, beta and gamma) on the effectiveness of the filters and on the robustness of the segmentation in the presence of qualitative variations of the CT scans. The parametric combinations generated through DOE were applied on a dataset of abdominal CT scans from the Medical Segmentation Decathlon, an open source collection containing abdominal images with segmentations of vessels and tumors. The segmentation results, obtained after the application of each parametric configuration, were compared with the corresponding masks of liver vessels using similarity metrics, such as the Dice Coefficient. The segmented models were subsequently exported in STL format for future holographic visualization, in order to provide advanced support to medical personnel for pre-operative planning.| File | Dimensione | Formato | |
|---|---|---|---|
|
Pinto_Ylenia.pdf
embargo fino al 06/10/2026
Dimensione
4.98 MB
Formato
Adobe PDF
|
4.98 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/93674