Spectral Bone Marrow

The Spectral Bone Marrow software utilizes deep learning for automated bone segmentation from spectral CT data to generate optimized visualizations of bone marrow, specifically by creating colored Water(HAP) material density images overlaid on base CT scans. This automated image processing tool is designed to enhance diagnostic capabilities for traumatic and non-traumatic bone pathologies while improving radiologist reader efficiency.

Spectral Bone Marrow is an automated image processing software application developed by GE Medical Systems, designed to enhance the visualization of bone marrow from spectral CT acquisitions. It employs deep learning technology for automatic bone segmentation. This segmented bone region is then utilized to create material density images, specifically water minus hydroxyapatite (Water(HAP)) images. These Water(HAP) images are subsequently colored and overlaid onto base monochromatic CT images or Virtual Unenhanced (VUE) images. The system is intended to assist clinicians in reviewing spectral body and extremity CT images for both traumatic and non-traumatic bone pathologies by providing an optimized view of the bone marrow. The software provides a fully automated post-processing workflow, aiming to improve reader efficiency. It offers features such as optional automatic setting of window width and level for the color overlay images and multiplanar export of the fused DICOM images. Spectral Bone Marrow is hosted on GE’s Edison Health Link (EHL) computational platform. Non-clinical engineering bench testing on 146 retrospective Spectral CT series, with ground truth established by radiologists, demonstrated the algorithm's capability for accurate bone segmentation. Clinical testing, involving retrospective assessment of clinical sample images by three board-certified radiologists, validated that the software provides additional diagnostic value for bone marrow evaluation and increases overall reader efficiency.

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 deep learning algorithm's performance in bone segmentation vary across different CT scanner models or acquisition protocols not included in the 146 retrospective series used for engineering bench testing?; What specific types of traumatic and non-traumatic bone pathologies showed the most significant improvement in diagnostic assessment using the Spectral Bone Marrow output compared to standard CT images or the predicate device?; Could the Water(HAP) material density values generated by the software be quantified and correlated with specific bone marrow conditions, such as cellularity changes or infiltration by disease, for more objective assessments?; What was the range of improvement in reader efficiency (e.g., time saved, diagnostic confidence) observed in the clinical testing, and were there specific user experience factors or case complexities where the efficiency gains were most pronounced?; How does the integration with GE’s Edison Health Link (EHL) computational platform impact the software's deployment, data processing times, and interoperability within diverse hospital IT infrastructures?