ABMD Software

The ABMD Software is an AI-driven, post-processing tool designed to opportunistically measure bone mineral density from existing CT scans by analyzing Hounsfield Units in vertebral trabecular bone. Its methodology involves AI-based segmentation and HU averaging to calculate BMD, T-scores, and Z-scores, providing a non-invasive method for assessing bone health without additional radiation exposure.

The ABMD Software is an AI-powered, post-processing tool designed to opportunistically measure bone mineral density (BMD) from existing CT scans of the vertebral bones. Its primary purpose is to enable retrospective BMD assessment, allow BMD evaluation during CT scans ordered for other medical reasons, or facilitate phantom-less BMD measurement, all without subjecting patients to additional radiation beyond their initial CT scan; the software analyzes DICOM CT images, extracts patient and scan metadata, and utilizes an AI model to segment vertebral bones. Following segmentation, the ABMD Software measures the average Hounsfield Units (HU) within a cylindrical volume of the trabecular tissue in each vertebral bone, ensuring exclusion of cortical bone, and the mean of these HU averages is then used to calculate BMD; subsequently, it computes Z-scores and T-scores using patient age, gender, and a CT scanner-specific calibration factor, outputting these results along with images for radiologist review. Performance studies demonstrated a strong correlation between ABMD Software and manual QCT-based BMD measurements (r=0.97), and a modest but significant correlation with DXA scans (r=0.72), aligning with literature comparisons of manual QCT and DXA and highlighting its specificity for trabecular bone assessment.

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 AI model's performance in vertebral bone segmentation vary across different CT scanner manufacturers, models, and image acquisition protocols?; What is the range of CT scanner-specific calibration factors, and how sensitive are the final BMD, T-score, and Z-score outputs to variations or inaccuracies in these input calibration factors?; Given the software relies on existing CT scans, what are the minimum image quality or specific scan parameters (e.g., slice thickness, reconstruction kernel) required for reliable ABMD measurements?; How does the software's performance compare to manual QCT when vertebral pathologies (e.g., fractures, metastases, degenerative changes) are present within the analyzed segments?; What is the clinical impact of the 'modest' correlation (r=0.72) with DXA, particularly in terms of diagnostic concordance for conditions like osteoporosis or osteopenia when using ABMD Software as an opportunistic tool?