Osteoporosis progresses gradually and painlessly over many years. Many people do not have their bone density measured until they break a bone or develop back pain. By then, bone loss is usually already well advanced. In the BMDNow joint research project funded by Hessen Agentur, researchers at the Fraunhofer Institute for Computer Graphics Research IGD are collaborating closely with Goethe University Frankfurt and IT company garritz online media international to develop software that helps detect osteoporosis at an early stage, i.e., before fractures occur. What sets the BMDNow software apart is its ability to use existing CT scans to precisely determine the bone mineral density (BMD) of individual vertebrae, with no need for additional staff or specialized equipment. The software uses an algorithm that combines neural networks for automatic image segmentation with advanced imaging methods to provide reliable bone density measurements.
Bone density analysis as a byproduct
Several imaging techniques are already available for diagnosing osteoporosis, the gold standard being dual-energy X-ray absorptiometry (DXA). However, this test is not typically performed until bone loss is already suspected. “In contrast, our software uses existing routine CT image data—such as from heart or lung scans—without the need for additional specialized examinations. This reduces patients’ individual radiation exposure and saves physicians’ time,” says Stefan Wesarg, a research scientist at Fraunhofer IGD in Darmstadt. “We run our algorithm on the available images and obtain a bone density analysis essentially as a byproduct.”
Preventing premature loss of mobility
But how does this software enable automated, precise bone density calculations? It first segments the trabecular region of the vertebral bodies, which is the spongy tissue inside the bone. To this end, the team tested various AI-based models under real-world conditions using actual patient data from Frankfurt University Hospital. “Our algorithm is applied specifically in this inner region of the vertebrae to generate a three-dimensional map of bone density distribution. The AI model analyzes bone components such as fat and calcium and uses color coding to visualize their proportions,” explains Wesarg. Bone density is calculated based on specific calibration curves previously generated using reference phantom models. The resulting values indicate whether osteoporosis or osteopenia—a precursor to osteoporosis—is present, enabling early initiation of the necessary treatment.
Second place in the Hessen Ideen competition
The researchers are currently using dual-energy CT scans, which scan the body at two different X-ray energy levels, to precisely measure the bone density of individual vertebral bodies. However, the project partners also plan to apply their algorithms to conventional CT scans (single-energy CT) in the future. The long-term goal is to establish a reference database, and the software is being trained and optimized accordingly. The project partners are currently finalizing and validating the software prototype, with medical device approval targeted within one to two years. The team has already won an award for its innovative software: The project took second place in the Hessen Ideen competition in 2025. The jury was particularly impressed by the project’s potential to reduce radiation exposure for individual patients and to improve efficiency in the healthcare system.
Fraunhofer-Gesellschaft
