Automatic Trajectory Planning from Preoperative MRI images Using the AI-Optimized RRT* Algorithm for Robotic Breast Ultrasound

Abstract

Ultrasound imaging is the most widely used technique for detecting and diagnosing breast cancer, as it provides real-time images in a non-invasive and inexpensive manner. However, performing ultrasound examinations requires experienced and well-trained operators either for probe manipulation during acquisition or during image interpretation, which makes this technique dependent. This dependency requires the employment of sonographers with extensive training and experience, which may limit access to this technique in remote areas and also expose experts to serious health issues due to repetitive movements. To overcome these difficulties, robotic ultrasound systems have emerged in recent years to enable conventional ultrasound examinations to be performed in an automated and standardized way. Our study contributes to this field through an integrated approach to automated planning of robotic breast ultrasound trajectories. This approach is based on the application of a developed version of the rapidly-exploring random tree (RRT) trajectory planning algorithm to pre-operational threedimensional (3D) magnetic resonance imaging (MRI) images. We performed simulations with the three versions of the algorithm RRT, RRT* and RRT* with 30% reinforcement learning (RL) on two different patients. The results obtained demonstrate the potential of the RRT algorithm with 30% RL to obtain optimized trajectories with high fluidity and a short path while ensuring total and homogeneous coverage adapted to different complex breast morphologies.