Abstract:Many imaging scenarios involve a sequence of tomographic data acquisitions to monitor change over time - e.g., longitudinal studies of disease progression (tumor surveillance) and intraoperative imaging of tissue changes during intervention. Radiation dose imparted for these repeat acquisitions present a concern. Because such image sequences share a great deal of information between acquisitions, using prior image information from baseline scans in the reconstruction of subsequent scans can relax data fidelity requirements of follow-up acquisitions. For example, sparse data acquisitions, including angular undersampling and limited-angle tomography, limit exposure by reducing the number of acquired projections. Various approaches such as prior-image constrained compressed sensing (PICCS) have successfully incorporated prior images in the reconstruction of such sparse data. Another technique to limit radiation dose is to reduce the x-ray fluence per projection. However, many methods for reconstruction of sparse data do not include a noise model accounting for stochastic fluctuations in such low-dose measurements and cannot balance the differing information content of various measurements. In this paper, we present a prior-image, penalized-likelihood estimator (PI-PLE) that utilizes prior image information, compressed-sensing penalties, and a Poisson noise model for measurements. The approach is applied to a lung nodule surveillance scenario with sparse data acquired at low exposures to illustrate performance under cases of extremely limited data fidelity. The results show that PI-PLE is able to greatly reduce streak artifacts that otherwise arise from photon starvation, and maintain high-resolution anatomical features, whereas traditional approaches are subject to streak artifacts or lower-resolution images.

Skellam Distribution Based Adaptive Two-Stage Non-Local Methods for Photon-Limited Poisson Noisy Image Reconstruction

Non-Local Image Inpainting Using Low-Rank Matrix Completion: Non-Local Image Inpainting Using Low-Rank Matrix Completion

Non-Lambertian photometric stereo by color and noise analysis

Nonlocal Low-Rank and Total Variation Constrained PET Image Reconstruction

Photon-Efficient 3D Imaging with A Non-local Neural Network

Hybrid High-Order Nonlocal Gradient Sparsity Regularization for Poisson Image Deconvolution.

Efficient L1/2 Regularization-Based Reconstruction for Photoacoustic Imaging Using Adaptively Iterative Thresholding Algorithm

Boosting Photon-Efficient Image Reconstruction With A Unified Deep Neural Network

Nonlocal Prior Bayesian Tomographic Reconstruction

Simultaneous Nonlocal Low-Rank and Deep Priors for Poisson Denoising

Reconstruction Algorithm of Low-Light Integral Imaging by Electron-Multiplying Charge-Coupled Device

Image Reconstruction with Locally Adaptive Sparsity and Nonlocal Robust Regularization

PET image bayesian reconstruction based on nonlocal steering kernel prior

Photon-efficient 3D Reconstruction Employing a Edge Enhancement Method.

Photon-Efficient Non-Line-of-Sight Imaging

An Adaptive Photon Count Reconstruction Algorithm for Sparse Count and Strong Noise Count Data With Low Signal Background Ratio

Photon-limited imaging through scattering medium based on deep learning

Photon Counting Integral Imaging Using Compound Photon Counting Model and Adaptive Parametric Maximum Likelihood Estimator

Incorporation of Noise and Prior Images in Penalized-Likelihood Reconstruction of Sparse Data

High-quality Reconstruction in Non-Line-of-sight Imaging Via Transient Multiplied Photon Count Data

Three dimensional photon counting integral imaging based on Bayesian adaptive reconstruction