Automatic Registration

The Automatic Registration device is engineered to improve image-guided surgical procedures by ensuring accurate, real-time alignment of intraoperatively acquired 3D CT/CBCT or fluoroscopic patient images with the surgical environment. It utilizes optical tracking and advanced software, now enhanced with a static AI/ML algorithm (Convolutional Neural Network) for automated vertebrae detection in 2D X-rays, to boost registration efficiency with specific imaging systems while upholding stringent accuracy standards.

The "Automatic Registration" device, version 2.6, is a software accessory for Brainlab's image-guided surgery navigation systems, specifically designed for systems like the Brainlab Spine & Trauma Navigation System. Its fundamental purpose is to achieve precise image registration by correlating intraoperatively acquired 3D CT, Cone-Beam CT (CBCT), or fluoroscopic images with the patient's anatomy within the surgical field, which is essential for accurate surgical navigation. The system integrates software modules including Automatic Registration 2.6, Universal Atlas Performer 6.0, and Universal Atlas Transfer Performer 6.0, along with hardware components such as registration matrices, adhesive flat markers, and a calibration phantom. It supports dedicated workflows for Airo and LoopX scanners, as well as a universal DICOM-based workflow compatible with other CT scanners. A significant technological advancement in version 2.6 is the 'Registration Update' method, which refines registrations initially generated with the LoopX scanner. This method incorporates a static (locked) Artificial Intelligence/Machine Learning (AI/ML) algorithm, specifically a Convolutional Neural Network (CNN) developed via supervised learning, sourced from the Universal Atlas. This AI/ML component automates the detection of vertebrae in 2D X-ray images, streamlining the pre-positioning process relative to the 3D dataset and thereby reducing the need for manual user adjustments. Performance evaluations, including bench tests on human cadavers, confirmed that this new AI-assisted registration method maintains the established accuracy benchmark of ≤ 2.5 mm (P95) with a mean 3D deviation of ≤ 1.5 mm. Furthermore, usability studies involving 15 representative users validated the effectiveness of the new user interface and workflow for the AI-enhanced feature, identifying no critical use-related problems.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of the AI/ML-based vertebrae detection vary across different patient anatomies (e.g., scoliosis, osteoporosis, previous fusions) or varying image qualities?; What are the specific failure modes of the 'Registration Update' AI/ML algorithm, and what manual override or correction mechanisms are in place for surgeons if misregistration occurs?; Beyond vertebrae, are there plans to extend the AI/ML capabilities to other anatomical landmarks or for different types of surgical procedures supported by Brainlab navigation systems?; What was the diversity of the dataset used to train the Convolutional Neural Network for vertebrae detection, and how does the 'locked' nature of the algorithm accommodate future improvements or address potential biases?; How does the reduced user interaction for pre-positioning, facilitated by the AI, translate into quantifiable benefits such as time savings or reduced cognitive load for the surgical team in a real-world operative setting?