Researchers from the Korea University College of Medicine have developed a groundbreaking tool called DL-BMD to revolutionize the screening of osteoporosis. This tool utilizes deep learning technology to automate the measurement of bone mineral density (BMD) on routine computed tomography (CT) scans, offering a more accessible and accurate approach to diagnosing osteoporosis.
Traditionally, the diagnosis of osteoporosis relied on central dual-energy X-ray absorptiometry (DXA). However, this method has limitations that contribute to the underdiagnosis and undertreatment of the condition. With DL-BMD, the need for specialized imaging techniques is eliminated, as the tool leverages existing CT scans for opportunistic osteoporosis screening.
To create DL-BMD, the researchers employed advanced deep-learning algorithms, building the tool upon a segmentation network called U-Net. Additional techniques, such as field of view augmentation and CT denoising, were incorporated to enhance reliability across different scan settings. The model was trained using a diverse dataset of CT scans, and various pre-processing steps and data augmentation were applied to improve its ability to generalize.
DL-BMD utilizes a region of interest (ROI) placement algorithm to create an elliptical ROI that excludes cortical bone and avoids the basivertebral vein. The lumbar vertebrae undergo calculation of Hounsfield unit (HU) values within the ROI. These values are then converted into BMD using calibration against the European Spine Phantom, ensuring accurate and reliable measurements.
This automated tool represents a significant advancement in osteoporosis screening, offering a highly efficient and precise solution. By addressing the limitations of traditional approaches, DL-BMD enables more accessible opportunistic screening through routine CT scans. This breakthrough holds promising potential for the early identification and proactive prevention of osteoporotic fractures, ultimately improving overall bone health on a larger scale.
The source of the article is from the blog crasel.tk