Biomedical Researchers at Geisel School of Medicine Achieve 80% Performance Increase When Training a Deep Learning System to Detect Osteoporosis

In a recent study published in Computers in Biology and Medicine, the Hassanpour Lab at Geisel School of Medicine reports findings on training the ResNet34 feature extraction network on a high-performance computer that was equipped with an NVIDIA Titan Xp GPU to detect osteoporotic vertebral fractures on CT scans. This innovative deep learning model matches the performance of practicing radiologists and can be used for screening and prioritizing fracture cases.