RESUMO
PURPOSE: To construct and evaluate the efficacy of a deep learning system to rapidly and automatically locate six vertebral landmarks, which are used to measure vertebral body heights, and to output spine angle measurements (lumbar lordosis angles [LLAs]) across multiple modalities. MATERIALS AND METHODS: In this retrospective study, MR (n = 1123), CT (n = 137), and radiographic (n = 484) images were used from a wide variety of patient populations, ages, disease stages, bone densities, and interventions (n = 1744 total patients, 64 years ± 8, 76.8% women; images acquired 2005-2020). Trained annotators assessed images and generated data necessary for deformity analysis and for model development. A neural network model was then trained to output vertebral body landmarks for vertebral height measurement. The network was trained and validated on 898 MR, 110 CT, and 387 radiographic images and was then evaluated or tested on the remaining images for measuring deformities and LLAs. The Pearson correlation coefficient was used in reporting LLA measurements. RESULTS: On the holdout testing dataset (225 MR, 27 CT, and 97 radiographic images), the network was able to measure vertebral heights (mean height percentage of error ± 1 standard deviation: MR images, 1.5% ± 0.3; CT scans, 1.9% ± 0.2; radiographs, 1.7% ± 0.4) and produce other measures such as the LLA (mean absolute error: MR images, 2.90°; CT scans, 2.26°; radiographs, 3.60°) in less than 1.7 seconds across MR, CT, and radiographic imaging studies. CONCLUSION: The developed network was able to rapidly measure morphometric quantities in vertebral bodies and output LLAs across multiple modalities.Keywords: Computer Aided Diagnosis (CAD), MRI, CT, Spine, Demineralization-Bone, Feature Detection Supplemental material is available for this article. © RSNA, 2021.
RESUMO
PURPOSE: Fractures in vertebral bodies are among the most common complications of osteoporosis and other bone diseases. However, studies that aim to predict future fractures and assess general spine health must manually delineate vertebral bodies and intervertebral discs in imaging studies for further radiomic analysis. This study aims to develop a deep learning system that can automatically and rapidly segment (delineate) vertebrae and discs in MR, CT, and X-ray imaging studies. RESULTS: We constructed a neural network to output 2D segmentations for MR, CT, and X-ray imaging studies. We trained the network on 4490 MR, 550 CT, and 1935 X-ray imaging studies (post-data augmentation) spanning a wide variety of patient populations, bone disease statuses, and ages from 2005 to 2020. Evaluated using 5-fold cross validation, the network was able to produce median Dice scores > 0.95 across all modalities for vertebral bodies and intervertebral discs (on the most central slice for MR/CT and on image for X-ray). Furthermore, radiomic features (skewness, kurtosis, mean of positive value pixels, and entropy) calculated from predicted segmentation masks were highly accurate (r ≥ 0.96 across all radiomic features when compared to ground truth). Mean time to produce outputs was <1.7 s across all modalities. CONCLUSIONS: Our network was able to rapidly produce segmentations for vertebral bodies and intervertebral discs for MR, CT, and X-ray imaging studies. Furthermore, radiomic quantities derived from these segmentations were highly accurate. Since this network produced outputs rapidly for these modalities which are commonly used, it can be put to immediate use for radiomic and clinical imaging studies assessing spine health.
