Solving ill-posed inverse problems using iterative deep neural networks

We propose a partially learned approach for the solution of ill posed inverse problems with not necessarily linear forward operators. The method builds on ideas from classical regularization theory and recent advances in deep learning to perform learning while making use of prior information about the inverse problem encoded in the forward operator, noise model and a regularizing functional. The method results in a gradient-like iterative scheme, where the "gradient" component is learned using a convolutional network that includes the gradients of the data discrepancy and regularizer as input in each iteration. We present results of such a partially learned gradient scheme on a non-linear tomographic inversion problem with simulated data from both the Sheep-Logan phantom as well as a head CT. The outcome is compared against FBP and TV reconstruction and the proposed method provides a 5.4 dB PSNR improvement over the TV reconstruction while being significantly faster, giving reconstructions of 512 x 512 volumes in about 0.4 seconds using a single GPU.

[1]  Alfred K. Louis,et al.  Feature reconstruction in inverse problems , 2011 .

[2]  Marcin Andrychowicz,et al.  Learning to learn by gradient descent by gradient descent , 2016, NIPS.

[3]  Otmar Scherzer,et al.  Shape Reconstruction with A Priori Knowledge Based on Integral Invariants , 2012, SIAM J. Imaging Sci..

[4]  Carola-Bibiane Schonlieb,et al.  Image denoising: learning noise distribution via PDE-constrained optimization , 2012, 1207.3425.

[5]  C. Schönlieb,et al.  Image denoising: Learning the noise model via nonsmooth PDE-constrained optimization , 2013 .

[6]  Nevzat Onur Domaniç,et al.  Shape-based image reconstruction using linearized deformations , 2017, Inverse problems.

[7]  Geoffrey E. Hinton,et al.  Deep Learning , 2015, Nature.

[8]  Kees Joost Batenburg,et al.  DART: A Practical Reconstruction Algorithm for Discrete Tomography , 2011, IEEE Transactions on Image Processing.

[9]  Ronny Ramlau,et al.  A Mumford-Shah level-set approach for the inversion and segmentation of X-ray tomography data , 2007, J. Comput. Phys..

[10]  Jian Sun,et al.  ADMM-Net: A Deep Learning Approach for Compressive Sensing MRI , 2017, ArXiv.

[11]  Stephen P. Boyd,et al.  Dirty Pixels: Optimizing Image Classification Architectures for Raw Sensor Data , 2017, ArXiv.

[12]  Thomas Schuster,et al.  The Method of Approximate Inverse: Theory and Applications , 2007 .

[13]  Chun-Liang Li,et al.  One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[14]  Lea Fleischer,et al.  Regularization of Inverse Problems , 1996 .

[15]  Carola-Bibiane Schönlieb,et al.  Bilevel Parameter Learning for Higher-Order Total Variation Regularisation Models , 2015, Journal of Mathematical Imaging and Vision.

[16]  Michael Unser,et al.  Deep Convolutional Neural Network for Inverse Problems in Imaging , 2016, IEEE Transactions on Image Processing.

[17]  Antonin Chambolle,et al.  A First-Order Primal-Dual Algorithm for Convex Problems with Applications to Imaging , 2011, Journal of Mathematical Imaging and Vision.

[18]  Andrew M. Stuart,et al.  Inverse problems: A Bayesian perspective , 2010, Acta Numerica.

[19]  Jan Sijbers,et al.  Fast and flexible X-ray tomography using the ASTRA toolbox. , 2016, Optics express.

[20]  Stanley Osher,et al.  A Guide to the TV Zoo , 2013 .

[21]  Jian Sun,et al.  Deep ADMM-Net for Compressive Sensing MRI , 2016, NIPS.

[22]  Carola-Bibiane Schönlieb,et al.  Bilevel approaches for learning of variational imaging models , 2015, ArXiv.

[23]  Jong Chul Ye,et al.  A deep convolutional neural network using directional wavelets for low‐dose X‐ray CT reconstruction , 2016, Medical physics.

[24]  Jorge Nocedal,et al.  On the limited memory BFGS method for large scale optimization , 1989, Math. Program..

[25]  Karl Kunisch,et al.  A Bilevel Optimization Approach for Parameter Learning in Variational Models , 2013, SIAM J. Imaging Sci..

[26]  J. C. De los Reyes,et al.  Learning optimal spatially-dependent regularization parameters in total variation image denoising , 2016, 1603.09155.

[27]  Andreas K. Maier,et al.  A Deep Learning Architecture for Limited-Angle Computed Tomography Reconstruction , 2017, Bildverarbeitung für die Medizin.

[28]  Kees Joost Batenburg,et al.  Fast Tomographic Reconstruction From Limited Data Using Artificial Neural Networks , 2013, IEEE Transactions on Image Processing.

[29]  Andreas K. Maier,et al.  Deep Learning Computed Tomography , 2016, MICCAI.

[30]  Sriram Subramaniam,et al.  Shape-Based Regularization of Electron Tomographic Reconstruction , 2012, IEEE Transactions on Medical Imaging.

[31]  Barbara Kaltenbacher,et al.  Iterative Regularization Methods for Nonlinear Ill-Posed Problems , 2008, Radon Series on Computational and Applied Mathematics.

[32]  Geoffrey E. Hinton,et al.  Rectified Linear Units Improve Restricted Boltzmann Machines , 2010, ICML.

[33]  K. Perez.  Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment , 2014 .

[34]  Max Welling,et al.  Recurrent Inference Machines for Solving Inverse Problems , 2017, ArXiv.

[35]  Carola-Bibiane Schönlieb,et al.  The structure of optimal parameters for image restoration problems , 2015, ArXiv.

[36]  P. Hansen.  Rank-Deficient and Discrete Ill-Posed Problems: Numerical Aspects of Linear Inversion , 1987 .

[37]  M. Bertero,et al.  Iterative image reconstruction : a point of view , 2007 .

[38]  Maria Argyrou,et al.  Tomographic Image Reconstruction based on Artificial Neural Network (ANN) techniques , 2012, 2012 IEEE Nuclear Science Symposium and Medical Imaging Conference Record (NSS/MIC).

[39]  N. Giokaris,et al.  Tomographic image reconstruction using Artificial Neural Networks , 2004 .

[40]  William H. Guss.  Deep Function Machines: Generalized Neural Networks for Topological Layer Expression , 2016, ArXiv.

[41]  Tuomo Valkonen,et al.  A primal–dual hybrid gradient method for nonlinear operators with applications to MRI , 2013, 1309.5032.