Ce travail reprend les études d’évaluation d’une plateforme instrumentale dédiée à la chirurgie mitrale mini-invasive vidéo-assistée, et du télémanipulateur Da Vinci pour la réalisation d’anastomose mammaire interne / IVA à thorax fermé.) Evaluation du Portaclamp. Cette étude clinique a porté sur 20 patients opérés de chirurgie cardiaque sous CEC et a confirmé la simplicité d’utilisation du système, son efficacité et l’absence de morbidité ou complication induite. L’étude chez le porc des effets histologiques sur la paroi de l’aorte des trois clamps (l’endo-clamp, le clamp Chitwood et le Portaclamp) a révélé une atteinte majeure de l’endothélium aortique induite par l’endo-clamp . 2) Evaluation du Portapleg. Le Portapleg est un dispositif auto-suturant de cardioplégie antérograde constitué d’un clip en Nitinol restant implanté sur l’aorte. Une étude sur 20 patients a été rapportée avec comme critère principal le temps de saignement du site de ponction après injection de protamine. Le système a montré son efficacité hémostatique dans tous les cas sans événement secondaire. 3) Evaluation du Mitrax’s. C’est un cône en plastique polymère auto ajustable et auto expansible, qui repousse de façon symétrique et concentrique les parois de l’oreillette. Une étude prospective de son efficacité a été réalisée chez 62 patients opérés de chirurgie mitrale vidéo-assistée de façon consécutive. L’indice de satisfaction a été en moyenne 4.6, témoin d’une exposition optimale de la valve mitrale avec une vision endoscopique ou directe de la valve mitrale jugée excellente. 4) Analyse comparative des techniques mini-invasives de revascularisation de l’IVA par pontage mammaire (Port Access, MIDCAB, TECAB). Cette étude prospective a porté sur 160 patients ; à trois mois, le taux de réintervention sur l’IVA était : PA-CABG, 0% ; MIDCAB, 1.8% ; TECAB, 10% ; p<0.01. A trois ans, les taux actuariels de survie sans réintervention étaient : PA-CABG, 100% ; MIDCAB, 98±5 % ; TECAB, 88±8 % ; p<0.05. / This thesis is based on the evaluation studies of an instrumental platform dedicated to video assisted minimally invasive mitral valve surgery, and of the robotic Da Vinci system in LAD bypass with mammary artery in a closed chest approach. 1) Evaluation of Portaclamp. In 20 patients who underwent cardiac surgery with Portaclamp, a clinical study showed that the clamping system is safe, fast and easy and does not generate undue morbidity. In a pig model, severe lesions of the intima were observed on the clamping spot with the endoclamp, in comparison with Portaclamp and Chitwood clamp. 2) Evaluation of Portapleg. Portapleg is an auto-suturing system dedicated to antegrade cardioplegia delivery, and based on a Nitinol clip left implanted on the aorta. In 20 patients, the closure of the puncture aortic hole and the haemostasis after protamine were obtained in all cases. The procedure did not generate undue morbidity and there was no device-related adverse event. 3) Evaluation of Mitrax’s. The Mitrax’s retractor is a pattern cut polymer sheet, self-expanding and auto-adjusting. The effectiveness of Mitrax’s was evaluated in 62 patients who consecutively underwent a video-assisted mitral valve procedure. The global satisfaction index was 4.6±0.5, demonstrating the effectiveness of the device which provides optimal exposure and excellent direct vision. 4) Comparative analysis of minimally invasive techniques for LAD revascularization with mammary artery graft (Port Access, MIDCAB, TECAB). In a prospective study, 160 patients were included. At 3-month postoperatively, the end-point of LAD reintervention were PA-CABG, 0%; MIDCAB, 1.8%; TECAB, 10%; p=0.01. At 3-year, reintervention-free survival was significantly lower in the TECAB group: PA-CABG, 100% ; MIDCAB, 98±5 % ; TECAB, 88±8 % ; p<0.05.

Background: Minimally invasive surgery (MIS) requires intense education and training with structured supervision and feedback. However, a standardized training structure is lacking in Germany. This nationwide survey aimed to assess the current state of minimally invasive surgery (MIS) training and factors impacting surgeons' satisfaction. - Methods: Between July and October 2021, an online survey was conducted among general, abdominal, and thoracic surgeons in Germany. The survey collected data on department size, individual operative experience, availability of MIS training equipment and curricula, and individual satisfaction with training. A linear regression analysis was conducted to investigate factors influencing the surgeons’ satisfaction with the MIS training. - Results: A total of 1008 surgeons participated in the survey, including residents (26.1%), fellows (14.6%), attendings (43.8%), and heads of departments (15.2%). Of the respondents, 57.4% reported having access to MIS training equipment, 29.8% and 26% had a curriculum for skills lab MIS training and intraoperative MIS training, respectively. In multivariate linear regression analysis, strongest predictors for surgeons’ satisfaction with skills lab MIS training and intraoperative training were the availability of respective training curricula (skills lab: β 12.572; p < 0.001 & intraoperative: β 16.541; p < 0.001), and equipment (β 5.246; p = 0.012 & β 4.295; p = 0.037), and experience as a first surgeon in laparoscopy (β 12.572; p < 0.001 & β 3.748; p = 0.007). Additionally, trainees and teachers differed in their satisfaction factors. - Conclusion: Germany lacks standardized training curricula and sufficient access to MIS training equipment. Trainees and teachers have distinct factors influencing their satisfaction with MIS training. Standardized curricula, equipment accessibility, and surgical experience are crucial for improving surgeons’ satisfaction with training.

<p dir="ltr">The field of reinforcement learning has made significant progress in recent years, with deep reinforcement learning (RL) being a major contributor. However, there are still challenges associated with the effective training of RL algorithms, particularly with respect to sample efficiency and generalization. This thesis aims to address these challenges by developing RL algorithms capable of generalizing to unseen environments and adapting to dynamic conditions, thereby expanding the practical applicability of RL in real-world tasks. The first contribution of this thesis is the development of novel policy optimization techniques that enhance the generalization capabilities of RL agents. These techniques include the Thinker method, which employs style transfer to diversify observation trajectories, and Bootstrap Advantage Estimation, which improves policy and value function learning through augmented data. These methods have demonstrated superior performance in standard benchmarks, outperforming existing data augmentation and policy optimization techniques. Additionally, this thesis introduces Robust Policy Optimization, a method that enhances exploration in policy gradient-based RL by perturbing action distributions. This method addresses the limitations of traditional methods, such as entropy collapse and primacy bias, resulting in improved sample efficiency and adaptability in continuous action spaces. The thesis further explores the potential of natural language descriptions as an alternative to image-based state representations in RL. This approach enhances interpretability and generalization in tasks involving complex visual observations by leveraging large language models. Furthermore, this work contributes to the field of semi-autonomous teleoperated robotic surgery by developing systems capable of performing complex surgical tasks remotely, even under challenging conditions such as communication delays and data scarcity. The creation of the DESK dataset supports knowledge transfer across different robotic platforms, further enhancing the capabilities of these systems. Overall, the advancements presented in this thesis represent significant steps toward developing more robust, adaptable, and efficient autonomous agents. These contributions have broad implications for various real-world applications, including autonomous systems, robotics, and safety-critical tasks such as medical surgery.</p>

This thesis outlines the development of an automatic image registration algorithm for matching 3D CT data to 2D fluoroscope X-ray images. The registration is required in order to calculate a transformation for measurements in the 2D image into the 3D representation. The algorithm achieves the registration by generating digitally reconstructed radiographs from the CT data set. The radiographs are 2D projection images, and therefore may be compared with the 2D Fluoroscope images. The X-ray and fluoroscope images were compared using the photometric-based registration algorithm, pseudocorrelation, with X^2 as the distance metric. An automated search algorithm was implemented using the Downhill Simplex of Nelder and Meade. The algorithm was successful in locating the position and orientation of the CT data set for calculating a digitally reconstructed radiograph to match the fluoroscope image. The CT data set was located with a maximum mean position error of 2.4 mm in xy, 4.4 mm in z, and xyz axial rotation within 0.5°. The standard deviation given 1800 random starting locations was 9.3 mm in x, 12.7 mm in y, 16.9 mm in z, xz axial rotation 2.5°, and y axial rotation of 1.9°. The search algorithm was successful in handling gross misalignment, however there were difficulties in convergence once within the vicinity of the global minimum. It is suggested to implement a hybrid search technique, switching to a conjugate gradient search algorithm once in the vicinity of the global minimum. An additional refinement would be a possible change of the distant metric, or the registration algorithm, once within the vicinity of the global minimum. Additional investigation needs to be directed towards testing the algorithm with live fluoroscope and CT data. This is required in order to assess registration performance when comparing different imaging modalities. / Thesis / Master of Engineering (ME)