Robot-assisted flexible needle insertion using universal distributional deep reinforcement learning.

PURPOSE: Flexible needle insertion is an important minimally invasive surgery approach for biopsy and radio-frequency ablation. This approach can minimize intraoperative trauma and improve postoperative recovery. We propose a new path planning framework using multi-goal deep reinforcement learning to overcome the difficulties in uncertain needle-tissue interactions and enhance the robustness of robot-assisted insertion process. METHODS: This framework utilizes a new algorithm called universal distributional Q-learning (UDQL) to learn a stable steering policy and perform risk management by visualizing the learned Q-value distribution. To further improve the robustness, universal value function approximation is leveraged in the training process of UDQL to maximize generalization and connect to diagnosis by adapting fast re-planning and transfer learning. RESULTS: Computer simulation and phantom experimental results show our proposed framework can securely steer flexible needles with high insertion accuracy and robustness. The framework also improves robustness by providing distribution information to clinicians for diagnosis and decision making during surgery. CONCLUSIONS: Compared with previous methods, the proposed framework can perform multi-target needle insertion through single insertion point qunder continuous state space model with higher accuracy and robustness.

Robust path planning for flexible needle insertion using Markov decision processes.

PURPOSE: Flexible needle has the potential to accurately navigate to a treatment region in the least invasive manner. We propose a new planning method using Markov decision processes (MDPs) for flexible needle navigation that can perform robust path planning and steering under the circumstance of complex tissue-needle interactions. METHODS: This method enhances the robustness of flexible needle steering from three different perspectives. First, the method considers the problem caused by soft tissue deformation. The method then resolves the common needle penetration failure caused by patterns of targets, while the last solution addresses the uncertainty issues in flexible needle motion due to complex and unpredictable tissue-needle interaction. RESULTS: Computer simulation and phantom experimental results show that the proposed method can perform robust planning and generate a secure control policy for flexible needle steering. Compared with a traditional method using MDPs, the proposed method achieves higher accuracy and probability of success in avoiding obstacles under complicated and uncertain tissue-needle interactions. Future work will involve experiment with biological tissue in vivo. CONCLUSION: The proposed robust path planning method can securely steer flexible needle within soft phantom tissues and achieve high adaptability in computer simulation.

Biprism distortion modeling and calibration for a single-lens stereovision system.

Biprism-based single-lens stereovision systems have several advantages over conventional two-camera systems, which has made them popular in recent years. However, employing a biprism in front of the camera will induce additional unique distortion in the image, which cannot be adequately represented by the existing distortion models, making it is difficult to calibrate or correct. In this work, a parametric biprism distortion model aiming to correct this biprism distortion is developed and evaluated. The estimation of the model starts with a general bivariate polynomial and is then refined based on distortion properties. These properties are determined by the geometrical analysis of the distortion formation in the system. This refined model is evaluated in two ways: distortion map data fitting and virtual camera calibration. Both simulation and actual experiments are carried out to show the feasibility of the proposed distortion model and the improvements in calibration accuracy and depth recovery compared with the existing distortion models.

Parameter error analysis of single-lens prism-based stereovision system.

In this paper, the quantization error and parameters of the single-lens prism-based stereovision system are studied in detail. Each image captured using this system can be divided into two subimages on the left and right, and these subimages are generated by two virtual cameras that are produced by the biprism. This stereovision system is equivalent to the conventional two-camera system and the two subimages captured have disparity that can be used to reconstruct the 3D scene. The relative range error of this system is studied with respect to the system parameters, which gives us a better way to quantify the range resolution of the system and enables the selection of system parameters to become easier.

Single-lens stereovision system using a prism: position estimation of a multi-ocular prism.

In this paper, a position estimation method using a prism-based single-lens stereovision system is proposed. A multifaced prism was considered as a single optical system composed of few refractive planes. A transformation matrix which relates the coordinates of an object point to its coordinates on the image plane through the refraction of the prism was derived based on geometrical optics. A mathematical model which is able to denote the position of an arbitrary faces prism with only seven parameters is introduced. This model further extends the application of the single-lens stereovision system using a prism to other areas. Experimentation results are presented to prove the effectiveness and robustness of our proposed model.

Projection-based visual guidance for robot-aided RF needle insertion.

PURPOSE: The use of projector-based augmented reality (AR) in surgery may enable surgeons to directly view anatomical models and surgical data from the patient's surface (skin). It has the advantages of a consistent viewing focus on the patient, an extended field of view and augmented interaction. This paper presents an AR guidance mechanism with a projector-camera system to provide the surgeon with direct visual feedback for supervision of robotic needle insertion in radiofrequency (RF) ablation treatment. METHODS: The registration of target organ models to specific positions on the patient body is performed using a surface-matching algorithm and point-based registration. An algorithm based on the extended Kalman filter and spatial transformation is used to intraoperatively compute the virtual needle's depth in the patient's body for AR display. RESULTS: Experiments of this AR system on a mannequin were conducted to evaluate AR visualization and accuracy of virtual RF needle insertion. The average accuracy of 1.86 mm for virtual needle insertion met the clinical requirement of 2 mm or better. The feasibility of augmented interaction with a surgical robot using the proposed open AR interface with active visual feedback was demonstrated. CONCLUSIONS: The experimental results demonstrate that this guidance system is effective in assisting a surgeon to perform a robot-assisted radiofrequency ablation procedure. The novelty of the work lies in establishing a navigational procedure for percutaneous surgical augmented intervention integrating a projection-based AR guidance and robotic implementation for surgical needle insertion.

Accurate geometrical optics model for single-lens stereovision system using a prism.

In this paper, we proposed a new method for analyzing the image formation of a prism. The prism was considered as a single optical system composed of some planes. By analyzing each plane individually and then combining them together, we derived a transformation matrix which can express the relationship between an object point and its image by the refraction of a prism. We also explained how to use this matrix for epipolar geometry and three-dimensional point reconstruction. Our method is based on optical geometry and could be used in a multiocular prism. Experimentation results are presented to prove the accuracy of our method is better than former researchers' and is comparable with that of the multicamera stereovision system.