Noise Reduction to Compute Tissue Mineral Density and Trabecular Bone Volume Fraction from Low Resolution QCT

Micro-structural parameters of the thoracic or lumbar spine generally carry insufficient accuracy and precision for clinical in vivo studies when assessed on quantitative computed tomography (QCT). We propose a 3D convolutional neural network with specific loss functions for QCT noise reduction to compute micro-structural parameters such as tissue mineral density (TMD) and bone volume ratio (BV/TV) with significantly higher accuracy than using no or standard noise reduction filters. The vertebra-phantom study contained high resolution peripheral and clinical CT scans with simulated in vivo CT noise and nine repetitions of three different tube currents (100, 250 and 360 mAs). Five-fold cross validation was performed on 20466 purely spongy pairs of noisy and ground-truth patches. Comparison of training and test errors revealed high robustness against over-fitting. While not showing effects for the assessment of BMD and voxel-wise densities, the filter improved thoroughly the computation of TMD and BV/TV with respect to the unfiltered data. Root-mean-square and accuracy errors of low resolution TMD and BV/TV decreased to less than 17% of the initial values. Furthermore filtered low resolution scans revealed still more TMD- and BV/TV-relevant information than high resolution CT scans, either unfiltered or filtered with two state-of-the-art standard denoising methods. The proposed architecture is threshold and rotational invariant, applicable on a wide range of image resolutions at once, and likely serves for an accurate computation of further micro-structural parameters. Furthermore, it is less prone for over-fitting than neural networks that compute structural parameters directly. In conclusion, the method is potentially important for the diagnosis of osteoporosis and other bone diseases since it allows to assess relevant 3D micro-structural information from standard low exposure CT protocols such as 100 mAs and 120 kVp.

[1]  Feng Lin,et al.  Low-Dose CT With a Residual Encoder-Decoder Convolutional Neural Network , 2017, IEEE Transactions on Medical Imaging.

[2]  Ronald M. Summers,et al.  Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning , 2016, IEEE Transactions on Medical Imaging.

[3]  Max A. Viergever,et al.  Generative Adversarial Networks for Noise Reduction in Low-Dose CT , 2017, IEEE Transactions on Medical Imaging.

[4]  P. Zysset,et al.  Comparative Effects of Teriparatide and Risedronate in Glucocorticoid-Induced Osteoporosis in Men: 18-Month Results of the EuroGIOPs Trial , 2013, Journal of bone and mineral research : the official journal of the American Society for Bone and Mineral Research.

[5]  Y. Lu,et al.  The role of patient-mode high-resolution peripheral quantitative computed tomography indices in the prediction of failure strength of the elderly women’s thoracic vertebral body , 2014, Osteoporosis International.

[6]  M. Heller,et al.  Association of QCT Bone Mineral Density and Bone Structure With Vertebral Fractures in Patients With Multiple Myeloma , 2015, Journal of bone and mineral research : the official journal of the American Society for Bone and Mineral Research.

[7]  Hu Chen,et al.  Low-dose CT via convolutional neural network. , 2017, Biomedical optics express.

[8]  Claudio Delrieux,et al.  Local texture descriptors for the assessment of differences in diffusion magnetic resonance imaging of the brain , 2017, International Journal of Computer Assisted Radiology and Surgery.

[9]  K. Engelke,et al.  High resolution computed tomography of the vertebrae yields accurate information on trabecular distances if processed by 3D fuzzy segmentation approaches. , 2009, Bone.

[10]  Alena-Kathrin Schnurr,et al.  Synthesis of CT images from digital body phantoms using CycleGAN , 2019, International Journal of Computer Assisted Radiology and Surgery.

[11]  Eli Konen,et al.  Creating Artificial Images for Radiology Applications Using Generative Adversarial Networks (GANs) - A Systematic Review. , 2020, Academic radiology.

[12]  M. Jergas,et al.  Accurate assessment of precision errors: How to measure the reproducibility of bone densitometry techniques , 2005, Osteoporosis International.

[13]  Eero P. Simoncelli,et al.  Image quality assessment: from error visibility to structural similarity , 2004, IEEE Transactions on Image Processing.

[14]  G. Huber,et al.  A new algorithm for estimating the rod volume fraction and the trabecular thickness from in vivo computed tomography. , 2016, Medical physics.

[15]  Rolando González-José,et al.  Automatic feature extraction and classification of Iberian ceramics based on deep convolutional networks , 2020, Journal of Cultural Heritage.

[16]  Örjan Smedby,et al.  Estimation of trabecular thickness in gray-scale images through granulometric analysis , 2012, Medical Imaging.

[17]  Emmanuel Iarussi,et al.  Generative Modelling of 3D in-silico Spongiosa with Controllable Micro-Structural Parameters , 2020, MICCAI.

[18]  A. Darabi,et al.  Thickness analysis and reconstruction of trabecular bone and bone substitute microstructure based on fuzzy distance map using both ridge and thinning skeletonization , 2009, Canadian Journal of Electrical and Computer Engineering.

[19]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[20]  Mannudeep K. Kalra,et al.  Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN) , 2017, ArXiv.

[21]  T. Ryan,et al.  Quantification and visualization of anisotropy in trabecular bone , 2004, Journal of microscopy.

[22]  Ge Wang,et al.  Structurally-Sensitive Multi-Scale Deep Neural Network for Low-Dose CT Denoising , 2018, IEEE Access.