The Development of an Autonomous Robotic Surgical Framework for Breast Brachytherapy

Abstract

Low-dose-rate-permanent-seed (LDR-PS) brachytherapy is a minimally inva- sive radiotherapy technique used after breast lumpectomy to prevent the re- growth of cancerous cells around the margins of a hollowed-out tumor (seroma). This approach involves implanting multiple radioactive seeds (each measuring 2-3mm in length) in and around the seroma, gradually irradiating and elimi- nating any remaining cancerous cells. LDR-PS brachytherapy has had signif- icant success in treating prostate cancer and is now being explored for breast cancer treatment. However, it has a more established history in the former. During LDR-PS, the seeds are implanted into the breast using 6-20 fine flexible needles under ultrasound (US) imaging, following a pre-operative plan derived from dosimetry calculations based on the patient’s medical images (usually Computed Tomography or CT). Besides its clinical benefits over other radiotherapy methods like external beam radiation, LDR-PS promotes health- care equity and inclusion by reducing frequent hospital visits, benefiting both rural and urban patients. However, the adoption of LDR-PS brachytherapy has encountered chal- lenges. One significant limitation is the requirement for surgeons to undergo substantial training for accurate seed implantation, particularly in breast can- cer cases. Inaccurate placement of seeds during breast brachytherapy is pri- marily due to two factors: (1) the discrepancy between the intraoperative ultrasound images and the pre-operative CT/MRI images caused by breast tissue deformation, and (2) the utilization of non-specialized surgical tools and techniques designed for prostate surgery. Incorrect seed implantation can lead to inadequate radiotherapy and increased cancer recurrence risk. Ad- ditionally, using instruments intended for prostate surgery is inappropriate due to the breast’s mobility and compliance, as well as the needle’s limited maneuverability due to its shorter insertion length. This study explores the potential benefits of incorporating an Assistive Robotic Surgical System (ARSS) in the context of LDR-PS brachytherapy surgery. The research focuses on addressing the complexities of the surgical environment, which undergoes deformation due to surgical interactions and necessitates patient-specific tuning. The study presents a comprehensive ap- proach to developing a simulation environment suitable for ARSS, beginning with pre-operative design and demonstrating its effectiveness in active defor- mation control during LDR-PS brachytherapy surgery. Moreover, the study investigates methods for updating the pre-operative model intra-operatively and explores the use of a robotic arm for US-probe manipulation. The re- search provides valuable insights into the potential applications of ARSS in enhancing the performance of LDR-PS brachytherapy surgery. The main con- tributions of this thesis are as follows: Active tissue deformation for target manipulation: This research tackles the lack of real-time nonlinear tissue modeling integrated into a control frame- work for tissue manipulation. The study demonstrates the integration of a real-time deformable tissue solver into the control loop, enabling effective tar- get manipulation. Intra-operative model updates for target tracking: Two methods are de- veloped to improve the accuracy of target tracking based on patient-specific biomechanical models. The first method, KF-ADMM, incorporates data into an ADMM-based Finite Element Method (FEM) solver through Kalman Fil- tering. The second method utilizes a generative variational autoencoder struc- ture based on graph neural networks (GNN-VAE) to reduce the dimensionality of the input mesh. The Ensemble Smoother with Multiple Data Assimilation (ES-MDA) is employed for simultaneous updates, enhancing the corrective ca- pability of the KF-ADMM method. Robot-assisted US probe manipulation: The study utilizes a Panda dexterous robotic arm to control the US probe, accurately following the needle tip. Overall, this research advances the field of ARSS in LDR-PS brachytherapy surgery by addressing tissue manipulation, target tracking, sim-to-real regis- tration, and robot-assisted US probe manipulation. The findings highlight the potential to improve the performance and precision of LDR-PS brachytherapy procedures, particularly in complex surgical environments.