Guided ultrasound acquisition for nonrigid image registration using reinforcement learning

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Machine Learning in Medicine

Background:

• Accurate spatial alignment of preoperative and intraoperative imaging is crucial for image-guided interventions.
• Ultrasound (US) imaging is widely used but often lacks precise spatial tracking, posing challenges for registration.
• Existing registration methods often struggle with the dynamic and heterogeneous nature of surgical environments.

Purpose of the Study:

• To develop and evaluate a novel guided registration framework for aligning preoperative and untracked ultrasound image slices.
• To leverage interactive and spatially adaptive techniques for optimizing ultrasound image acquisition during registration.
• To demonstrate the efficacy of proactive image acquisition in improving nonrigid registration accuracy in simulated surgical interventions.

Main Methods:

• A framework combining a deep hyper-network-based registration function and a reinforcement learning (RL) function for image acquisition guidance.
• The RL function estimates optimal image acquisition locations and adapts deformation regularization based on acquisition position.
• The method was evaluated using real preoperative patient data and simulated intraoperative data with varying fields of view.

Main Results:

• The guided registration method demonstrated statistically significant improvements in overall registration performance across multiple metrics compared to baseline methods.
• Significant enhancements were observed compared to registration without acquisition guidance, adaptable deformation regularization, classical iterative methods, and existing learning-based registration techniques.
• The study provides the first demonstration of proactive image acquisition's efficacy in simulated surgical interventional registration.

Conclusions:

• The proposed guided registration framework effectively improves the spatial alignment of preoperative and ultrasound images.
• Proactive, guided image acquisition is a viable strategy to overcome limitations in nonrigid registration for surgical interventions.
• This approach offers a promising direction for enhancing the accuracy and reliability of image-guided surgery.