Motion planning for a three-stage multilumen transoral lung access system

Abstract

Lung cancer is the leading cause of cancer-related death, and early-stage diagnosis is critical to survival. Biopsy is typically required for a definitive diagnosis, but current low-risk clinical options for lung biopsy cannot access all biopsy sites. We introduce a motion planner for a multilumen transoral lung access system, a new system that has the potential to perform safe biopsies anywhere in the lung, which could enable more effective early-stage diagnosis of lung cancer. The system consists of three stages in which a bronchoscope is deployed transorally to the lung, a concentric tube robot pierces through the bronchial tubes into the lung parenchyma, and a steerable needle deploys through a properly oriented concentric tube and steers through the lung parenchyma to the target site while avoiding anatomical obstacles such as significant blood vessels. A sampling-based motion planner computes actions for each stage of the system and considers the coupling of the stages in an efficient manner. We demonstrate the motion planner's fast performance and ability to compute plans with high clearance from obstacles in simulated anatomical scenarios.

Decoupling Steerability From Diameter: Helical Dovetail Laser Patterning for Steerable Needles

“…Using magnetic tracker measurements collected during insertion, we measured the radius of curvature in inflated lung as 100 mm. Note that the best prior radius of curvature result we have achieved in deflated lung tissue was 255 mm [5], [9]. Inflation reduces lung density, making needles steer with worse curvature in inflated lung.…”

GA3C Reinforcement Learning for Surgical Steerable Catheter Path Planning

“…Aleksandra et al [22] integrated sampling-based path planning with reinforcement learning agents for indoor navigation and aerial cargo delivery. Learning based methods, and in particular reinforcement learning, are more flexible approaches with respect to graphbased and sampling-based methods, as those presented by [24], [25], [26], allowing one to directly include all the expected features (obstacle clearance, kinematic constraints meeting, minimum trajectory length) in the optimization process, without the need of subsequent refinement steps, which are time-consuming and may still not provide the optimal trajectory. The goal of this work is to explore a reinforcement learning approach in the context of KN, to overcome the above mentioned limitation of classical and modified versions of sampling-based and graph-based methods, when dealing with steerable needles.…”

SLAM-Based Follow-the-Leader Deployment of Concentric Tube Robots

“…Traditionally, navigation can be achieved with two approaches, either open-loop based on an offline trajectory planning [7] using 3D models reconstructed from medical images [8], or using automatic navigation [9]. Implementation of image-based deployment relies on the use of an external imaging system for an eye-to-hand configuration, or on an eye-in-hand one when a visual sensor is mounted at the tip of the robot.…”