The left atrial appendage (LAA) closure is a crucial procedure to reduce the risk of thrombus formation in patients with atrial fibrillation, a condition that significantly increases the risk of embolic events and stroke. Minimally invasive transcatheter procedures, such as LAA closure, require high levels of skill from the operator, who must navigate precisely through delicate anatomical structures, often exposed to harmful radiation due to fluoroscopic guidance. The European ARTERY project aims to eliminate fluoroscopy by replacing it with automated catheter path planning, which can improve procedural precision and safety, while also reducing operative times. This study proposes a reinforcement learning approach, the Soft Actor-Critic (SAC) algorithm, to automate dynamic catheter path planning during LAA closure. The SAC algorithm was trained and validated in a dynamic digital twin of the cardiac anatomy, which models the real-time movement of the left atrium, with the atrial walls acting as dynamic obstacles. The algorithm receives real-time observations such as the distances between the catheter and the atrial walls and the catheter’s position relative to the target. The objective is to plan optimal paths that balance computational efficiency and path length. Additionally, it is crucial to ensure that the catheter approaches the LAA ostium as perpendicular as possible to facilitate device deployment, while also ensuring the paths are smooth to minimize collision risks. The performance of SAC was evaluated by comparing it with a Rapidly-exploring Random Tree (RRT) replanning algorithm and with manually generated paths created by an experienced cardiovascular surgeon. SAC demonstrated superior performance, achieving a 100% success rate and producing faster, more precise, and smoother trajectories. In particular, SAC significantly reduced trajectory times and minimized orientation errors compared to the other methods. These findings suggest that reinforcement learning can offer a promising solution for improving the automation and safety of robotic catheter navigation in dynamic cardiac environments.
La chiusura dell’auricola sinistra (LAA) è una procedura essenziale per ridurre il rischio di formazione di trombi nei pazienti affetti da fibrillazione atriale, una condizione che aumenta significativamente il rischio di eventi embolici e ictus. Le procedure percutanee minimamente invasive, come la chiusura dell’auricola, richiedono una grande abilità da parte dell’operatore, il quale deve navigare con precisione attraverso strutture anatomiche delicate, spesso esposto a radiazioni dannose a causa della guida tramite fluoroscopia. Il progetto europeo ARTERY, mira a eliminare la fluoroscopia sostituendola con una pianificazione automatizzata del percorso del catetere, che possa migliorare la precisione e la sicurezza delle procedure, oltre a ridurre i tempi operatori. Questo studio propone un approccio basato sull’algoritmo di apprendimento per rinforzo, Soft Actor-Critic (SAC), per automatizzare la pianificazione del percorso del catetere durante la chiusura della LAA in un ambiente dinamico che simula il movimento del cuore. Il SAC è stato addestrato e validato utilizzando un digital twin dinamico dell’anatomia cardiaca, capace di riprodurre in tempo reale il movimento dell’atrio sinistro, con le pareti atriali che agiscono come ostacoli dinamici. L’algoritmo riceve, in tempo reale, osservazione come le distanze tra il catetere e le pareti e la posizione relativa del catetere rispetto al target, con l’obiettivo di pianificare percorsi ottimali sia in termini di tempo di esecuzione che di lunghezza del tragitto. Un ulteriore obiettivo è garantire che il catetere raggiunga l’ostium dell’auricola in maniera quanto più possibile perpendicolare, così da facilitare l’impianto del dispositivo, e che il percorso sia "smooth" per minimizzare i rischi di collisione. I risultati ottenuti con SAC sono stati confrontati con un algoritmo di Rapidly-exploring Random Tree replanning e con percorsi generati manualmente da un chirurgo cardiovascolare. SAC ha dimostrato una performance superiore, raggiungendo il target con il 100% di successo e producendo traiettorie più rapide, precise e lineari. In particolare, l’algoritmo SAC ha ridotto significativamente i tempi di percorrenza e ha minimizzato gli errori di orientamento rispetto agli altri metodi. Questi risultati suggeriscono che l’approccio basato sull’apprendimento per rinforzo rappresenti una soluzione promettente per migliorare l’automazione e la sicurezza delle procedure di navigazione robotica del catetere in ambienti cardiaci dinamici.
Development of a path planning algorithm for Left Atrial Appendage Closure (LAAC) procedure
TRAVERSA, ALICE
2023/2024
Abstract
The left atrial appendage (LAA) closure is a crucial procedure to reduce the risk of thrombus formation in patients with atrial fibrillation, a condition that significantly increases the risk of embolic events and stroke. Minimally invasive transcatheter procedures, such as LAA closure, require high levels of skill from the operator, who must navigate precisely through delicate anatomical structures, often exposed to harmful radiation due to fluoroscopic guidance. The European ARTERY project aims to eliminate fluoroscopy by replacing it with automated catheter path planning, which can improve procedural precision and safety, while also reducing operative times. This study proposes a reinforcement learning approach, the Soft Actor-Critic (SAC) algorithm, to automate dynamic catheter path planning during LAA closure. The SAC algorithm was trained and validated in a dynamic digital twin of the cardiac anatomy, which models the real-time movement of the left atrium, with the atrial walls acting as dynamic obstacles. The algorithm receives real-time observations such as the distances between the catheter and the atrial walls and the catheter’s position relative to the target. The objective is to plan optimal paths that balance computational efficiency and path length. Additionally, it is crucial to ensure that the catheter approaches the LAA ostium as perpendicular as possible to facilitate device deployment, while also ensuring the paths are smooth to minimize collision risks. The performance of SAC was evaluated by comparing it with a Rapidly-exploring Random Tree (RRT) replanning algorithm and with manually generated paths created by an experienced cardiovascular surgeon. SAC demonstrated superior performance, achieving a 100% success rate and producing faster, more precise, and smoother trajectories. In particular, SAC significantly reduced trajectory times and minimized orientation errors compared to the other methods. These findings suggest that reinforcement learning can offer a promising solution for improving the automation and safety of robotic catheter navigation in dynamic cardiac environments.File | Dimensione | Formato | |
---|---|---|---|
2024_10_Traversa_ExecutiveSummary.pdf
solo utenti autorizzati a partire dal 17/09/2025
Descrizione: executive_summary
Dimensione
1.61 MB
Formato
Adobe PDF
|
1.61 MB | Adobe PDF | Visualizza/Apri |
2024_10_Traversa_Tesi.pdf
solo utenti autorizzati a partire dal 17/09/2025
Descrizione: testo tesi
Dimensione
17.5 MB
Formato
Adobe PDF
|
17.5 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/10589/226927