A precise segmentation of organs at risk (OARs) and of the Clinical Target Volume (CTV) is required by Radiotherapy (RT) to maximize treatment efficacy and, at the same time, minimize toxicity. Deep learning (DL) has done a lot of steps forward in automatic contouring, but complex targets like CTVs remain challenging. This thesis explores the use of simpler, well-segmented structures (e.g., OARs) as Anatomical Prior (AP) information and the division of the body into anatomical regions to improve CTV segmentation. We found that APs can improve the segmentation of the CTV only in certain conditions, as well as the division of the body into anatomical regions. The results suggest that the best approach is a hybrid strategy that combines the regional models and the global models.
Una segmentazione precisa degli organi a rischio (OAR) e del Volume Target Clinico (CTV) è fondamentale nella Radioterapia (RT) per massimizzare l'efficacia del trattamento e, allo stesso tempo, ridurre al minimo la tossicità. Il deep learning (DL) ha fatto grandi progressi nella contornazione automatica, ma la segmentazione di target complessi come i CTV rimane una sfida. Questa tesi esplora l'uso di strutture più semplici e ben segmentate (ad esempio, gli OAR) come informazioni anatomiche prioritarie (AP) e la suddivisione del corpo in regioni anatomiche per migliorare la segmentazione del CTV\@. Ciò che abbiamo scoperto è che gli AP possono migliorare la segmentazione del CTV solo in determinate condizioni, così come la suddivisione del paziente in regioni anatomiche. I risultati suggeriscono che l'approccio migliore è una strategia ibrida che combina i modelli regionali e quelli globali.
Enhancing medical imaging segmentation with anatomical priors
BATTISTINI, JODY ROBERTO
2023/2024
Abstract
A precise segmentation of organs at risk (OARs) and of the Clinical Target Volume (CTV) is required by Radiotherapy (RT) to maximize treatment efficacy and, at the same time, minimize toxicity. Deep learning (DL) has done a lot of steps forward in automatic contouring, but complex targets like CTVs remain challenging. This thesis explores the use of simpler, well-segmented structures (e.g., OARs) as Anatomical Prior (AP) information and the division of the body into anatomical regions to improve CTV segmentation. We found that APs can improve the segmentation of the CTV only in certain conditions, as well as the division of the body into anatomical regions. The results suggest that the best approach is a hybrid strategy that combines the regional models and the global models.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Battistini.pdf
accessibile in internet per tutti
Dimensione
5.91 MB
Formato
Adobe PDF
|
5.91 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/10589/234272