Session Item

Implementation of new technology, techniques, clinical protocols or trials (including QA & audit)
Digital Poster
Physics
CoBra: Towards Robotic Real Time MRI-Guided Prostate Brachytherapy and Biopsy
Sarah Wilby, United Kingdom
PO-1711

Abstract

CoBra: Towards Robotic Real Time MRI-Guided Prostate Brachytherapy and Biopsy
Authors:

Sarah Wilby1, Antony Palmer1, Wojciech Polak1, Sepaldeep Singh Dhaliwal2, Sina Firouzy3, Ashraf Labib4, Dylan Jones5, Stefan Escaida Navarro2, Rochdi Merzouki2, Kevin Boni Brou6, David Pasquier7,8, John van den Dobbelsteen 9, Martijn de Vries9

1Portsmouth Hospital University Trust, Medical Physics, Portsmouth, United Kingdom; 2University of Lille, CRIStAL, CNRS UMR 9189, Lille, France; 3University of Portsmouth, School of Business, Portsmouth, United Kingdom; 4University of Portsmouth, Business and Law, Portsmouth, United Kingdom; 5University of Portsmouth, School of Mathematics, Portsmouth, United Kingdom; 6Centre Oscar Lambret, Medical Physics, Lille, France; 7Centre Oscar Lambret, Academic Department of Radiation Oncology, Lille, France; 8University of Lille, CRIStAL, UMR 9189, Lille, France; 9TU Delft, Department of BioMechanical Engineering, Delft, The Netherlands

Show Affiliations
Purpose or Objective

Progress report on key scientific and technical developments from the multi-institution collaborative project, CoBra. Interreg 2 Seas (EU)-funded project (ends September 2022).

Material and Methods

Development of a novel, robotic prostate biopsy (BX) and brachytherapy (BT) device. The CoBra MRI-guided robot aims to perform adaptive BT under real-time (RT)-MRI, accounting for target changes due to tissue deformation.  The robot has 5 DoF, is actuated with ultrasonic–motors in closed-loop feedback and is MRI-compatible. Patient remains in-bore throughout. Curved needle insertion allows avoidance of OARs or obstructions. A quick-lock mechanism allows easy exchange of the BX or BT modules. Seed delivery adapts to the new target position during the implant, with the aid of omnidirectional steerable needles and RT dose calculations.

Use of RT-MR guided BT, requires rapid creation of accurate synthetic-CT (sCT) datasets from the live MR images.  We use an algorithm called augmented cycle Generative Adversarial Network (AugCGAN).  This is more robust with the variability of MR images than the standard cycleGAN.  This study included T2w MR and CT pelvic images of 38 patients from 5 centres. The AugCGAN was trained on 2D transverse slices of 19 patients from 3 different sites. The network was then used to generate sCT images of 19 patients coming from two other sites. Mean Absolute Errors (MAE) for each patient were evaluated between real and sCT.

The needle path planning algorithm is designed for the MR-teerable needles used in the CoBra project.  The algorithm receives the seed plan and prostate contours from MRI as an input. It clusters the seeds and creates candidate path plans to reach all seed positions, using a single insertion point. Fig. 1 is an example of four path candidates. Acceptable results are those that allow all seed positions to be reached, while keeping a minimum, pre-set distance between the needle and the urethra. From all candidates, the one that causes the least amount of tissue damage, is chosen.

For robot testing, an active bio-inspired prostate phantom (BIP), depicting the motion, deformation and inflammation of the prostate seen clinically during needle insertion, has been developed. The BIP is connected to SOFA (simulation open framework architecture) to estimate the prostate changes, thus enabling tracking of target position interactively (fig. 2). 



Fig. 1.  Needle trajectory planning software


Fig. 2. BIP in MRI scanner with CoBra robot
Results

Tests of the BIP and robotic needle insertion in RT-MRI, showed that the BIP features could be extracted with good fidelity and the needle tip identified under T1w, T2w and DYN-B FEE sequences.  sCT datasets showed a mean MAE of 59.8 HU and 65.8 HU for first and second test sites, taking on average 8.5 s to generate a complete sCT (90 slices) on our GPU (Nvidia Quadro P6000).

Conclusion

This report demonstrates some of the latest progress achieved from the CoBra project, as it reaches the phantom testing stage.