Pancreatic Ductal AdenoCarcinoma (PDAC) represents one of the most aggressive malignancies among pancreatic cancers, with a five-year survivability below 10\%. Its poor pronosis is mainly related to late detection and the challenges associated with accurately localizing lesions during diagnosis and intervention. To address these limitations and improve therapeutic outcomes, this work develops and validates a preliminary navigation support tool designed to enhance visualization of anatomical structures and improve the clinician's awareness of the probe pose during clinical intervention. The proposed framework fuses the information from the two common medical imaging modalities applied in suspected PDAC cases: the spatial accuracy and completeness of Computerized Tomography (CT) and the interactivity of the ultrasound exploration. The resulting system provides real-time tracking of probe position in the 3D CT volume through a unified and interactive visualization environment. More importantly, the proposed solution does not require any clinical annotation for its application, but automatically pre-process ultrasound videos and CT scans with hand crafted algorithms. Thus, it is suitable for application in smaller clinical settings where highly specialized personnel might not be available. In addition, this study provides novelty to the scientific literature as it represents the first study to validate unsupervised 3D probe alignment on real acquired ultrasound images. In this context, the discrepancy between ultrasound label and automatic CT segmentation at the predicted position and orientation are used as accuracy metric and demonstrates a relatively accurate CT-to-US registration and reliable tracking of the endoscopic probe trajectory. Moreover, the proposed algorithm is tested against other common solutions present in literature, outperforming them and achieving performance comparable to more advanced tools present in literature. The results show the potential applicability for the guidance in endoscopic interventions to detect lesions in improving lesion localization and diagnostic confidence. While only the validation is performed, the findings provides a promising direction for future research when clinical data is limited.
L'adenocarcinoma dei dotti pancreatici(PDAC) rappresenta uno dei più aggressivi tumori maligni tra le neoplasie del pancreas, con una soppravvivenza a cinque anni dalla diagnosi inferiore al 10\%. La prognosi infausta è dovuta soprattuto a ritardi nella diagnosi ed alla difficoltà nel localizzare le lesioni in sede diagnostica. Per superare queste limitazioni e migliorare il risultato degli interventi, questo lavoro presenta una prima implementazione di un sistema di supporto all'esplorazione con assistenza nella visualizzazione delle strutture biologiche e nel riconoscimento della posizione della sonda. Questo strumento combina le informazioni presenti nelle analisi effettuate nei casi di sospetto PDAC: l'accuratezza e completezza della Tomografia Assiale Computerizzata (TAC) con l'interattività dell'ecografia endoscopica. Il sistema risultante restituisce in tempo reale la posizione e l'orientazione della sonda nel volume 3D in un'interfaccia grafica unificata ed intuitiva. E soprattutto la soluzione proposta non richiede interventi diretti del clinico sui dati acquisiti, ma le immagini mediche sono processate algoritmi preimpostati. Di conseguenza, è molto efficace per strutture mediche relativamente piccole senza personale medico specializzato in manipolazione dei dati. In aggiunta, questo studio rappresenta il primo strumento in letteratura per l'allineamento di sonde endoscopiche nello spazio 3D senza supervisione umana e validato su ecografie reali. L'accuratezza dello strumento viene valutata calcolando la differenza tra la posizione delle strutture nell'ecografia e nella TAC alla posizione predetta. Questa procedura dimostra l'affidabilità e la sufficiente accuratezza dello strumento nel tracciamento in tempo reale della posizione della sonda. Inoltre, quando confrontato con altre soluzioni proposte in letteratura, dimostra performance superiori e paragonabili a strumenti sviluppati con marcata supervisione umana. I risultati riportati mostrano la rilevanza di questo sistema di guida in interventi endoscopici per migliorare il riconoscimento di lesioni la sicurezza della diagnosi. Per quanto la validazione effettuata sia limitata, questi risultati preliminari forniscono idee promettenti per la ricerca futura, in particolare con dati limitati.
Implementation and preliminary validation of an integrated navigation support tool for precise pancreatic cancer diagnosis
Grassi, Cristian
2025/2026
Abstract
Pancreatic Ductal AdenoCarcinoma (PDAC) represents one of the most aggressive malignancies among pancreatic cancers, with a five-year survivability below 10\%. Its poor pronosis is mainly related to late detection and the challenges associated with accurately localizing lesions during diagnosis and intervention. To address these limitations and improve therapeutic outcomes, this work develops and validates a preliminary navigation support tool designed to enhance visualization of anatomical structures and improve the clinician's awareness of the probe pose during clinical intervention. The proposed framework fuses the information from the two common medical imaging modalities applied in suspected PDAC cases: the spatial accuracy and completeness of Computerized Tomography (CT) and the interactivity of the ultrasound exploration. The resulting system provides real-time tracking of probe position in the 3D CT volume through a unified and interactive visualization environment. More importantly, the proposed solution does not require any clinical annotation for its application, but automatically pre-process ultrasound videos and CT scans with hand crafted algorithms. Thus, it is suitable for application in smaller clinical settings where highly specialized personnel might not be available. In addition, this study provides novelty to the scientific literature as it represents the first study to validate unsupervised 3D probe alignment on real acquired ultrasound images. In this context, the discrepancy between ultrasound label and automatic CT segmentation at the predicted position and orientation are used as accuracy metric and demonstrates a relatively accurate CT-to-US registration and reliable tracking of the endoscopic probe trajectory. Moreover, the proposed algorithm is tested against other common solutions present in literature, outperforming them and achieving performance comparable to more advanced tools present in literature. The results show the potential applicability for the guidance in endoscopic interventions to detect lesions in improving lesion localization and diagnostic confidence. While only the validation is performed, the findings provides a promising direction for future research when clinical data is limited.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Grassi_Tesi.pdf
non accessibile
Dimensione
33.88 MB
Formato
Adobe PDF
|
33.88 MB | Adobe PDF | Visualizza/Apri |
|
2025_12_Grassi_Executive_summary.pdf
non accessibile
Dimensione
736.73 kB
Formato
Adobe PDF
|
736.73 kB | 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/10589/247633