Design and development of estimation and control system for robot-assisted surgery.

Robot-assisted surgery represents a significant advancement in the medical field, addressing the limitations of traditional methods that are often influenced by human factors such as tremors and fatigue. These factors can lead to inconsistent outcomes and extended recovery times. Robotic systems improve precision, control, and visualization, particularly through minimally invasive techniques, which result in smaller incisions and shorter recovery periods. However, critical challenges remain in robot-assisted surgery (RAS). Significant key issues include accurately measuring spatial information related to surgical tools, detecting tools within a surgical context, maneuvering tools to avoid collisions, and consistently executing repetitive tasks such as cutting and stitching. Specifically, obtaining accurate spatial feedback regarding the position, angle, and distance of tools in cluttered environments is challenging. Moreover, detecting instruments in complex scenarios such as those involving blood and light reflections adds to the difficulty, especially on lower-end platforms with limited GPU capabilities. Ensuring precision while avoiding collisions during tool maneuvers and carrying out repetitive tasks in dynamically constrained environments demands considerable concentration. Furthermore, dependence on automated systems introduces risks in high-pressure situations that require quick decision-making. These challenges underscore the limitations of robot-assisted surgery and highlight the necessity for advanced research aimed at effectively automating complex surgical tasks. To address these issues, this research proposes multi-model integrated systems (frameworks), featuring a triple-layer mathematical model, an improved deep learning model, and a control-kinematics architecture. The first framework employs a deep learning model combined with a Kalman filter for continuous tracking and spatial feedback estimation using quadratic polynomials, Euclidean distance, and vector-based angle measurement. Additionally, an artificial potential field is used to ensure collision-free paths for trainee surgeons. Next, an improved deep learning model that detects surgical tools effectively in challenging environments. Additionally, it is integrated with model predictive control and a geometric-based analytical inverse kinematics solver to enable collision-free control in constrained and highly dynamic environments. Lastly, a novel control system-kinematic architecture is designed that involves a discrete Lyapunov-adapted linear quadratic regulator with a proportional integral derivative and an improved Levenberg-Marquardt solver. A machine learning model is also integrated for hybrid inverse kinematics within a graphical user interface-based human-robot interaction framework, facilitating efficient repetition of surgical tasks with rapid precision. Overall, the proposed approaches ensure virtual assistance for surgeons and enable the robotic system to perform autonomously with minimal user interaction. The proposed groundbreaking novel approaches addressed critically challenging tasks in robotic surgery by achieving notable results in precise spatial feedback estimation in both simulated and surgical contexts, as well as accurately detecting surgical tools in highly blurred and noisy conditions without compromising performance speed, especially in lower-end operating platforms. Additionally, it achieved remarkable results in planning an optimal path with reduced computation time, minimal reach error, and shorter path length. Similarly, it solved the inverse kinematics problem with excellent results, securing rapid and precise convergence. In summary, it demonstrated superior performance over conventional-to-recent state-of-the-art methods in terms of various metrics, such as spatial assessment accuracy, detection precision, collision-free planning, and controlling efficiency. The aim of these improvements is to make teleoperated and automated surgical systems more accurate and efficient, which will lead to better patient outcomes and the future of minimally invasive surgery.

Author dc:creator

• Muthukrishnan, Ramkumar

Advisor dc:contributor.advisor

• S. Kanna, M.S. Mekala and Y. Zhao

• Robot-assisted surgery
• Spatial feedback estimation
• Collision-free path navigation
• Deep learning
• Kalman filter
• Artificial potential field
• Model predictive control
• Geometric-based analytical inverse kinematic

Language dc:language

en

Identifier

oai:rgu-repository.worktribe.com:3476040
https://doi.org/10.48526/rgu-wt-3476040

Author Identifier

0009-0007-4598-5973

OAI identifier oai:identifier

oai:rgu-repository.worktribe.com:3476040