BoneMRI

BoneMRI is an advanced image processing software designed to improve the visualization of osseous structures from MRI scans by enhancing bone-to-soft-tissue contrast. It employs a convolutional neural network to analyze 3D gradient echo MRI data, generating synthetic CT-like images that provide quantitative insights into 3D bone morphology and tissue radiodensity for the pelvic and spinal regions.

BoneMRI is an image processing software designed to enhance the visualization of bone structures in MRI images by increasing their contrast against surrounding soft tissue. It is indicated for use in the pelvic region (including sacrum, hip bones, and femoral heads) and the spine (cervical, thoracic, lumbar, and S1 vertebrae) for patients aged 12 years and older. The software analyzes 3D gradient echo MRI scans, employing a convolutional neural network-based algorithm to construct 3D tomographic radiodensity contrast images. While not intended for tumor diagnosis or as a general replacement for CT scans, BoneMRI aids in visualizing 3D bone morphology, tissue radiodensity, and tissue radiodensity contrast. The algorithm is trained on diverse datasets from multiple clinical sites, anatomies, and scanners, assigning Hounsfield Unit (HU) values based on intensity and contextual information. BoneMRI can be deployed on-premise or as a managed cloud service, with data de-identified for cloud processing, and integrates with DICOM archives. Performance validation, conducted on retrospective clinical data from 193 patients (76 pelvic, 117 spine) by comparing BoneMRI images to co-registered CT scans, demonstrated clinically acceptable accuracy. Key performance metrics included mean absolute cortical delineation error below 1.0 mm, mean tissue radiodensity deviation below 25 HU (below 55 HU for bone), and a mean HU correlation coefficient above 0.75 for bone, consistently across various subgroups including age, geographical location, and BMI.

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 BoneMRI algorithm specifically differentiate cortical bone from trabecular bone, and what is its accuracy in representing subtle bone pathologies like microfractures or early osteonecrosis?; What are the specific parameters of the "dedicated MRI scan protocol" (3D gradient echo) required for optimal BoneMRI image generation, and how sensitive is the output to deviations from this protocol?; Beyond the listed anatomical regions (pelvis and spine), what are the potential technical limitations or challenges in extending BoneMRI's application to other areas with complex bony anatomy, such as extremities or the cranium?; How does the performance of BoneMRI, particularly in terms of radiodensity accuracy and cortical delineation, compare across different MRI field strengths (e.g., 1.5T vs. 3T) included in the validation?; Considering the algorithm is trained on datasets from multiple sites and scanners, what is the process for ensuring continued robustness and generalizability as new MRI hardware and software versions are introduced in clinical practice?