SIS System (Version 5.1.0)

The SIS System version 5.1.0 is a software application utilizing deep learning convolutional neural networks, trained on 7T MRI data, to generate patient-specific 3D anatomical models of brain structures like the STN and GPe/GPi from clinical MR images. Its purpose is to aid medical professionals in viewing, image processing, image fusion, and intraoperative functional planning, including the localization and orientation detection of implanted leads from CT scans.

The SIS System (version 5.1.0) is a software application designed for medical professionals, including surgeons, neurologists, and radiologists, to assist in the viewing, presentation, and documentation of medical imaging. It incorporates modules for image processing, image fusion, and intraoperative functional planning, with its 3D output being compatible with stereotactic image-guided surgery systems for further processing and visualization. The system specifically aids in the visualization of the subthalamic nuclei (STN) and globus pallidus externa and interna (GPe and GPi), supplementing standard clinical methods by providing additional adjunctive information for planning stereotactic surgical procedures and lead output. The system's technology is based on machine learning and image processing, utilizing pre-trained deep learning convolutional neural network (CNN) models to create patient-specific 3D anatomical models of these brain structures from a patient's clinical MR images. These CNN models were trained using ultra-high resolution 7T Magnetic Resonance images as ground truth. A key update in version 5.1.0 includes an algorithm based on these deep learning models, the addition of GPe/GPi as visualization targets, and a new functionality to determine the orientation of directional leads segmented from post-operative CT images. Performance data demonstrated that visualization accuracy for STN and GPi/GPe met acceptance criteria, MRI to CT registration accuracy was maintained, and lead segmentation from CT images was validated. The electrode orientation detection was accurate within ±30° in 91% of trusted detections, with over 90% of all presented orientations meeting this accuracy.

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 deep learning CNN models for segmenting GPe/GPi compare to existing gold-standard manual segmentation techniques in terms of accuracy and time efficiency?; What specific architectural features of the convolutional neural network models were optimized for brain image segmentation, and how do they contribute to improved visualization of STN and GPe/GPi?; What is the clinical impact of the ±30° accuracy range for directional lead orientation detection on postoperative programming and patient outcomes?; How does the system handle variability in MR image quality and acquisition parameters across different clinical sites when applying the pre-trained deep learning models?; Beyond STN and GPe/GPi, what other brain structures could this deep learning methodology be applied to for enhanced visualization and surgical planning?