Ultralow-parameter denoising: Trainable bilateral filter layers in computed tomography.


Journal

Medical physics
ISSN: 2473-4209
Titre abrégé: Med Phys
Pays: United States
ID NLM: 0425746

Informations de publication

Date de publication:
Aug 2022
Historique:
revised: 25 04 2022
received: 31 01 2022
accepted: 11 05 2022
pubmed: 19 5 2022
medline: 19 8 2022
entrez: 18 5 2022
Statut: ppublish

Résumé

Computed tomography (CT) is widely used as an imaging tool to visualize three-dimensional structures with expressive bone-soft tissue contrast. However, CT resolution can be severely degraded through low-dose acquisitions, highlighting the importance of effective denoising algorithms. Most data-driven denoising techniques are based on deep neural networks, and therefore, contain hundreds of thousands of trainable parameters, making them incomprehensible and prone to prediction failures. Developing understandable and robust denoising algorithms achieving state-of-the-art performance helps to minimize radiation dose while maintaining data integrity. This work presents an open-source CT denoising framework based on the idea of bilateral filtering. We propose a bilateral filter that can be incorporated into any deep learning pipeline and optimized in a purely data-driven way by calculating the gradient flow toward its hyperparameters and its input. Denoising in pure image-to-image pipelines and across different domains such as raw detector data and reconstructed volume, using a differentiable backprojection layer, is demonstrated. In contrast to other models, our bilateral filter layer consists of only four trainable parameters and constrains the applied operation to follow the traditional bilateral filter algorithm by design. Although only using three spatial parameters and one intensity range parameter per filter layer, the proposed denoising pipelines can compete with deep state-of-the-art denoising architectures with several hundred thousand parameters. Competitive denoising performance is achieved on x-ray microscope bone data and the 2016 Low Dose CT Grand Challenge data set. We report structural similarity index measures of 0.7094 and 0.9674 and peak signal-to-noise ratio values of 33.17 and 43.07 on the respective data sets. Due to the extremely low number of trainable parameters with well-defined effect, prediction reliance and data integrity is guaranteed at any time in the proposed pipelines, in contrast to most other deep learning-based denoising architectures.

Sections du résumé

BACKGROUND BACKGROUND
Computed tomography (CT) is widely used as an imaging tool to visualize three-dimensional structures with expressive bone-soft tissue contrast. However, CT resolution can be severely degraded through low-dose acquisitions, highlighting the importance of effective denoising algorithms.
PURPOSE OBJECTIVE
Most data-driven denoising techniques are based on deep neural networks, and therefore, contain hundreds of thousands of trainable parameters, making them incomprehensible and prone to prediction failures. Developing understandable and robust denoising algorithms achieving state-of-the-art performance helps to minimize radiation dose while maintaining data integrity.
METHODS METHODS
This work presents an open-source CT denoising framework based on the idea of bilateral filtering. We propose a bilateral filter that can be incorporated into any deep learning pipeline and optimized in a purely data-driven way by calculating the gradient flow toward its hyperparameters and its input. Denoising in pure image-to-image pipelines and across different domains such as raw detector data and reconstructed volume, using a differentiable backprojection layer, is demonstrated. In contrast to other models, our bilateral filter layer consists of only four trainable parameters and constrains the applied operation to follow the traditional bilateral filter algorithm by design.
RESULTS RESULTS
Although only using three spatial parameters and one intensity range parameter per filter layer, the proposed denoising pipelines can compete with deep state-of-the-art denoising architectures with several hundred thousand parameters. Competitive denoising performance is achieved on x-ray microscope bone data and the 2016 Low Dose CT Grand Challenge data set. We report structural similarity index measures of 0.7094 and 0.9674 and peak signal-to-noise ratio values of 33.17 and 43.07 on the respective data sets.
CONCLUSIONS CONCLUSIONS
Due to the extremely low number of trainable parameters with well-defined effect, prediction reliance and data integrity is guaranteed at any time in the proposed pipelines, in contrast to most other deep learning-based denoising architectures.

Identifiants

pubmed: 35583171
doi: 10.1002/mp.15718
doi:

Types de publication

Journal Article

Langues

eng

Sous-ensembles de citation

IM

Pagination

5107-5120

Subventions

Organisme : European Research Council
ID : 810316
Pays : International

Informations de copyright

© 2022 The Authors. Medical Physics published by Wiley Periodicals LLC on behalf of American Association of Physicists in Medicine.

Références

Boone JM, Hendee WR, McNitt-Gray MF, Seltzer SE. Radiation exposure from CT scans: how to close our knowledge gaps, monitor and safeguard exposure-proceedings and recommendations of the Radiation Dose Summit, sponsored by NIBIB, February 24-25, 2011. Radiology. 2012;265:544-554.
Barrett HH, Gordon S, Hershel R. Statistical limitations in transaxial tomography. Comput Biol Med. 1976;6:307-323.
Maier A, Fahrig R. GPU denoising for computed tomography. In: Graphics Processing Unit-Based High Performance Computing in Radiation Therapy. CRC Press; 2015;113.
Kelm ZS, Blezek D, Bartholmai B, Erickson BJ. Optimizing non-local means for denoising low dose CT. In: 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro. IEEE; 2009:662-665.
Maier A, Wigström L, Hofmann HG, et al. Three-dimensional anisotropic adaptive filtering of projection data for noise reduction in cone beam CT. Med Phys. 2011;38:5896-5909.
Manduca A, Yu L, Trzasko JD, et al. Projection space denoising with bilateral filtering and CT noise modeling for dose reduction in CT. Med Phys. 2009;36:4911-4919.
Manhart M, Fahrig R, Hornegger J, Doerfler A, Maier A. Guided noise reduction for spectral CT with energy-selective photon counting detectors. In: Proceedings of the Third CT Meeting. 2014:91-94.
Han X, Bian J, Eaker DR, et al. Algorithm-enabled low-dose micro-CT imaging. IEEE Trans Med Imaging. 2010;30:606-620.
Beister M, Kolditz D, Kalender WA. Iterative reconstruction methods in X-ray CT. Physica Med. 2012;28:94-108.
Gilbert P. Iterative methods for the three-dimensional reconstruction of an object from projections. J Theor Biol. 1972;36:105-117.
Chen H, Zhang Y, Kalra MK, et al. Low-dose CT with a residual encoder-decoder convolutional neural network. IEEE Trans Med Imaging. 2017;36:2524-2535.
Patwari M, Gutjahr R, Raupach R, Maier A. JBFnet - low dose CT denoising by trainable joint bilateral filtering. In: International Conference on Medical Image Computing and Computer-Assisted Intervention-MICCAI 2020. Springer; 2020:506-515.
Fan F, Shan H, Kalra MK, et al. Quadratic autoencoder (Q-AE) for low-dose CT denoising. IEEE Trans Med Imaging. 2019;39:2035-2050.
Yang Q, Yan P, Zhang Y, et al. Low-dose CT image denoising using a generative adversarial network with Wasserstein distance and perceptual loss. IEEE Trans Med Imaging. 2018;37:1348-1357.
Kang E, Chang W, Yoo J, Ye JC. Deep convolutional framelet denosing for low-dose CT via wavelet residual network. IEEE Trans Med Imaging. 2018;37:1358-1369.
Kang E, Min J, Ye JC. A deep convolutional neural network using directional wavelets for low-dose X-ray CT reconstruction. Med Phys. 2017;44:e360-e375.
Ketcha MD, Marrama M, Souza A, et al. Sinogram + image domain neural network approach for metal artifact reduction in low-dose cone-beam computed tomography. J Med Imaging. 2021;8:1-16.
Yuan X, He P, Zhu Q, Li X. Adversarial examples: attacks and defenses for deep learning. IEEE Trans Neural Netw Learn Syst. 2019;30:2805-2824.
Wu P, Sisniega A, Uneri A, et al. Using uncertainty in deep learning reconstruction for cone-beam CT of the brain. arXiv preprint arXiv:2108.09229. 2021.
Zhang C, Li Y, Chen G-H. Deep learning in image reconstruction: vulnerability under adversarial attacks and potential defense strategies. In: Medical Imaging 2021: Physics of Medical Imaging, Vol. 11595, International Society for Optics and Photonics; 2021:115951U.
Antun V, Renna F, Poon C, Adcock B, Hansen AC. On instabilities of deep learning in image reconstruction and the potential costs of AI. Proc Natl Acad Sci USA. 2020;117:30088-30095.
Huang Y, Würfl T, Breininger K, Liu L, Lauritsch G, Maier A. Some investigations on robustness of deep learning in limited angle tomography. In: International Conference on Medical Image Computing and Computer-Assisted Intervention-MICCAI 2018. Springer; 2018:145-153.
Tomasi C, Manduchi R. Bilateral filtering for gray and color images. In: Sixth International Conference on Computer Vision. IEEE; 1998:839-846.
Chen Y, Shu Y. Optimization of bilateral filter parameters via chi-square unbiased risk estimate. IEEE Signal Process Lett. 2013;21:97-100.
Anoop V, Bipin PR. Medical image enhancement by a bilateral filter using optimization technique. J Med Syst. 2019;43:1-12.
Kishan H, Seelamantula CS. Optimal parameter selection for bilateral filters using Poisson Unbiased Risk Estimate. In: 2012 19th IEEE International Conference on Image Processing. IEEE; 2012:121-124.
Dai T, Zhang Y, Dong L, Li L, Liu X, Xia S. Content-aware bilateral filtering. In: 2018 IEEE Fourth International Conference on Multimedia Big Data (BigMM). IEEE; 2018:1-6.
Peng H, Rao R. Bilateral kernel parameter optimization by risk minimization. In: 2010 IEEE International Conference on Image Processing. IEEE; 2010:3293-3296.
Jampani V, Kiefel M, Gehler PV. Learning sparse high dimensional filters: image filtering, dense crfs and bilateral neural networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016:4452-4461.
Kang W, Patwari M. Low Dose Helical CBCT denoising by using domain filtering with deep reinforcement learning. arXiv preprint arXiv:2104.00889. 2021.
Patwari M, Gutjahr R, Raupach R, Maier A. Limited parameter denoising for low-dose X-ray computed tomography using deep reinforcement learning. Med Phys. 2022.
Paszke A, Gross S, Massa F, et al. PyTorch: an imperative style, high-performance deep learning library. In: Proceedings of NeurIPS. Curran Associates, Inc.; 2019:8024-8035.
Syben C, Michen M, Stimpel B, Seitz S, Ploner S, Maier AK. PYRO-NN: python reconstruction operators in neural networks. Med Phys. 2019;46:5110-5115.
Krull A, Buchholz T-O, Jug F. Noise2void-learning denoising from single noisy images. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019:2129-2137.
Maier A, Syben C, Stimpel B, et al. Learning with known operators reduces maximum error bounds. Nat Machine Intell. 2019;1:373-380.
Maier A, Schebesch F, Syben C, et al. Precision learning: towards use of known operators in neural networks. In: 2018 24th International Conference on Pattern Recognition (ICPR). IEEE; 2018:183-188.
McCollough CH, Bartley AC, Carter RE, et al. Low-dose CT for the detection and classification of metastatic liver lesions: results of the 2016 low dose CT grand challenge. Med Phys. 2017;44:e339-e352.
Jakob W. pybind11 - Seamless operability between C++11 and Python. 2021.
Thies M. Differentiable reconstruction for x-ray microscopy data. 2021. https://doi.org/10.24433/CO.2740182.v2
Feldkamp LA, Davis LC, Kress JW. Practical cone-beam algorithm. Josa a. 1984;1:612-619.
Hannah KM, Thomas CD, Clement JG, De Carlo F, Peele AG. Bimodal distribution of osteocyte lacunar size in the human femoral cortex as revealed by micro-CT. Bone. 2010;47:866-871.
Grüneboom A, Hawwari I, Weidner D, et al. A network of trans-cortical capillaries as mainstay for blood circulation in long bones. Nat Metabol. 2019;1:236-250.
Grüneboom A, Kling L, Christiansen S, et al. Next-generation imaging of the skeletal system and its blood supply. Nat Rev Rheumatol. 2019;15:533-549.
Gruber R, Pietschmann P, Peterlik M. Introduction to bone development, remodelling and repair. In: Grampp S, ed. Radiology of Osteoporosis. Springer; 2008:1-23.
Buenzli PR, Sims NA. Quantifying the osteocyte network in the human skeleton. Bone. 2015;75:144-150.
Wagner F, Thies M, Karolczak M, et al. Monte Carlo dose simulation for in-vivo X-ray nanoscopy. In: Bildverarbeitung für die Medizin 2022. Springer; 2022:107-112.
Mill L, Kling L, Grüneboom A, Schett G, Christiansen S, Maier A. Towards in-vivo X-ray nanoscopy. In: Bildverarbeitung für die Medizin 2019. Springer; 2019:251-256.
Huang Y, Mill L, Stoll R, et al. Semi-permeable filters for interior region of interest dose reduction in X-ray microscopy. In: Bildverarbeitung für die Medizin. Springer; 2021:61-66.
Aust O, Thies M, Weidner D, et al. Tibia cortical bone segmentation in micro-CT and X-ray microscopy data using a single neural network. In: Bildverarbeitung für die Medizin 2022. Springer; 2022:333-338.
Yu L, Shiung M, Jondal D, McCollough CH. Development and validation of a practical lower-dose-simulation tool for optimizing computed tomography scan protocols. J Comput Assist Tomogr. 2012;36:477-487.
Barron AR. Approximation and estimation bounds for artificial neural networks. Mach Learn. 1994;14:115-133.
Moen TR, Chen B, Holmes III DR, et al. Low-dose CT image and projection dataset. Med Phys. 2021;48:902-911.

Auteurs

Fabian Wagner (F)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Mareike Thies (M)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Mingxuan Gu (M)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Yixing Huang (Y)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Sabrina Pechmann (S)

Fraunhofer Institute for Ceramic Technologies and Systems IKTS, Forchheim, 91301, Germany.

Mayank Patwari (M)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Stefan Ploner (S)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Oliver Aust (O)

Department of Internal Medicine 3 - Rheumatology and Immunology, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91054, Germany.
University Hospital Erlangen, Erlangen, 91054, Germany.

Stefan Uderhardt (S)

Department of Internal Medicine 3 - Rheumatology and Immunology, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91054, Germany.
University Hospital Erlangen, Erlangen, 91054, Germany.

Georg Schett (G)

Department of Internal Medicine 3 - Rheumatology and Immunology, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91054, Germany.
University Hospital Erlangen, Erlangen, 91054, Germany.

Silke Christiansen (S)

Fraunhofer Institute for Ceramic Technologies and Systems IKTS, Forchheim, 91301, Germany.
Institute for Nanotechnology and Correlative Microscopy e.V. INAM, Forchheim, 91301, Germany.

Andreas Maier (A)

Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, 91058, Germany.

Articles similaires

Selecting optimal software code descriptors-The case of Java.

Yegor Bugayenko, Zamira Kholmatova, Artem Kruglov et al.
1.00
Software Algorithms Programming Languages
Humans Magnetic Resonance Imaging Phantoms, Imaging Infant, Newborn Signal-To-Noise Ratio
1.00
Humans Magnetic Resonance Imaging Brain Infant, Newborn Infant, Premature
Cephalometry Humans Anatomic Landmarks Software Internet

Classifications MeSH