Robotic ultrasound-guided hybrid tracking framework for surgical interventions with respiratory movement

Ultrasound (US) is a technique widely used for its ease of use, real-time imaging and low cost, employed for diagnostics and minimally invasive procedures such as biopsies, targeted drug delivery and tumor ablation. This technology requires a lot of experience for excellent results and can stressful for both patients and the medical staff; moreover, respiration-induced motion, if not addressed properly, reduces the effectiveness of such procedures, prolongs treatment times and decreases patient comfort. Collaborative robotics can solve these issues and guarantee repeatability, accuracy and constant contact force applied to patients, but it does not always succeed in adequately compensating for respiration-induced motion. Traditionally, respiration-induced motion is handled either by patient breath-holding or with motion-compensation beds, both of which present limitations in terms of patient comfort, cost, and accuracy. This thesis examines the possibility of using an innovative technique for compensating respiratory motion without relying on additional supports or on practices that may be uncomfortable. In this work, a Convolutional Neural Network (CNN) is used to extract information within the image, followed by an Optical Flow (OF) algorithm to track motion and finally a Kalman Filter (KF) to predict future positions and provide compensation. The validation of the proposed framework was performed via multiple experiments at different respiration velocities, measuring the tracking error, defined as the distance between the target position and the measured position of the robot. The experimental results demonstrated that a Proportional Derivative (PD) controller ensured a tracking error below 2 mm. A needle insertion task was then carried out to assess its applicability in a simulated clinical scenario and a volunteer study was finally conducted to evaluate whether the controller, combined with a needle guide for assisted injections, could be deemed safe for the operators, thereby confirming the hypothesis and the validity of the proposed solution. This thesis examines the possibility of using an innovative technique to adequately compensate for respiratory motion without relying on additional supports such as motion-compensation beds or on practices that may be uncomfortable for patients, such as breath-holding. In this approach, a Convolutional Neural Network (CNN) is used to extract information about moving components within the image, followed by an Optical Flow (OF) algorithm to track their motion and finally a Kalman Filter (KF) to predict future positions and provide improved compensation. The technique was validated by measuring the delay between what was displayed in the image and the simulated respiratory motion. A controller was then selected to track the target within the image quickly and safely, ensuring a tracking error below 2 mm. Finally, a volunteer study was carried out to verify that this controller, together with a needle guide enabling controlled injections, was effective in improving comfort during the procedure, thereby confirming the hypothesis and the validity of the proposed solution.

Gli ultrasuoni sono molto usati in medicina per facilità d’uso, imaging in tempo reale e basso costo, impiegati sia nella diagnostica che in procedure mininvasive come biopsie, somministrazione di farmaci e ablazione tumorale. Questa tecnologia richiede esperienza per risultati ottimi e può essere stressante per pazienti e il personale; inoltre, ignorare il movimento respiratorio riduce l’efficacia delle procedure, aumenta i tempi di trattamento e diminuisce il comfort. La robotica collaborativa in questo caso può migliorare l'esperienza garantendo ripetibilità, accuratezza e forza di contatto costante, ma non sempre è in grado di compensare adeguatamente il movimento indotto dalla respirazione. Oggi questo problema viene affrontato chiedendo al paziente di trattenere il respiro o usando lettini a compensazione automatica, soluzioni limitate per comfort, costi e precisione. In questa tesi si esplora l’uso innovativo di un robot con una sonda a ultrasuoni attaccata al suo end-effector per compensare il movimento respiratorio senza supporti aggiuntivi né pratiche scomode per il paziente. L’approccio prevede una rete neurale convoluzionale per estrarre informazioni dall’immagine, un algoritmo di flusso ottico per seguirne il movimento e un filtro di Kalman per stimare la posizione futura del bersaglio e consentire una migliore compensazione. La validazione è stata effettuata tramite esperimenti a diverse velocità, misurando l'errore di tracciamento, ovvero la distanza tra la posizione del bersaglio e quella misurata del robot. I risultati hanno mostrato che un controllore Proporzionale Derivativo (PD) abbinato a questo metodo garantiva un errore di tracciamento inferiore a 2 mm. Infine, è stato eseguito un test di inserimento di un ago in uno scenario clinico simulato per verificare la sua applicabilità, e un test con dei volontari per valutare la sicurezza dell’approccio combinato con una guida per iniezioni assistite, confermando la validità della soluzione proposta.