Interior photon-counting computed tomography (PCCT) scans are essential for obtaining high-resolution images at minimal radiation dose by focusing only on a region of interest. However, designing a deep learning model for denoising a PCCT interior scan is rather challenging. Recently, several studies explored deep reinforcement learning (RL)-based models with far fewer parameters than those typical for supervised and self-learning models. Such an RL model can be effectively trained on a small dataset, and yet be generalizable and interpretable. In this work, we design an RL model to perform multichannel PCCT scan denoising. Because a reliable reward function is crucial for optimizing the RL model, we focus on designing a small denoising autoencoder-based reward network to learn the latent representation of full-dose simulated PCCT data and use the reconstruction error to quantify the reward. We also use domain-specific batch normalization for unsupervised domain adaptation with a limited amount of multichannel PCCT data. Our results show that the proposed model achieves excellent denoising results, with a significant potential for clinical and preclinical PCCT denoising.
Machine learning, especially convolutional neural network (CNN) approach has been successfully applied in noise suppression in natural image. However, shifting from natural image to medical image filed remains challenging due to specific difficulties such as training samples limitation, clinically meaningful image quality requirement and so on. To address this challenge, one possible solution is to incorporate our human prior knowledge into the machine learning model to better benefit its power. Therefore, in this work, we propose one prior knowledge driven machine learning based approach for positron emission tomography (PET) sinogram data denoising. Two main properties of PET sinogram data were considered in CNN architecture design, which are the Poisson statistics of the data and different correlation strength in the detector and view directions. Specially, for the statistical property, the sparse non-local method was used. For the correlation property, separate convolution was applied in two directions respectively. Experimental results showed the proposed model outperform the CNN model without prior knowledge. Results also demonstrate our insight of applying human knowledge strength the power of machine learning in medical imaging field.
Photon-counting spectral computed tomography (PCCT) reconstructs multiple energy-channel images to describe the same object, where there exists a strong correlation among all channels. In addition, reconstruction of each energychannel image suffers photon count starving problem. To make full use of the correlation among different channels to suppress the data noise and enhance the tissue texture in reconstructing each energy-channel image, this paper proposed a tensor convolutional neural network (TCNN) architecture to learn a tissue-specific texture prior for PCCT reconstruction. Specifically, we first model the spatial texture prior information in each individual channel using a convolution neural network, and then extract the correlation information among different energy channels by merging the multi-channel networks. Finally, we integrate the TCNN as a prior into Bayesian reconstruction framework. To evaluate the tissue texture preserving performance of the proposed method for each channel, a vivid clinical phantom which can simulate the real tissue textures was employed. The improvement associated with TCNN is remarkable relative to simultaneous algebraic reconstruction technique (SART) and tensor dictionary learning (TDL) based reconstruction. The proposed method produced promising results in terms of not only preserving texture feature but also suppressing image noise in each channel. The proposed method outperforms the competing methods in both visual inspection and quantitative indexes of root mean square error (RMSE), peak signal to noise ratio (PSNR), structural similarity (SSIM) and feature similarity (FSIM).
Purpose: Bayesian theory provides a sound framework for ultralow-dose computed tomography (ULdCT) image reconstruction with two terms for modeling the data statistical property and incorporating a priori knowledge for the image that is to be reconstructed. We investigate the feasibility of using a machine learning (ML) strategy, particularly the convolutional neural network (CNN), to construct a tissue-specific texture prior from previous full-dose computed tomography.
Approach: Our study constructs four tissue-specific texture priors, corresponding with lung, bone, fat, and muscle, and integrates the prior with the prelog shift Poisson (SP) data property for Bayesian reconstruction of ULdCT images. The Bayesian reconstruction was implemented by an algorithm called SP-CNN-T and compared with our previous Markov random field (MRF)-based tissue-specific texture prior algorithm called SP-MRF-T.
Results: In addition to conventional quantitative measures, mean squared error and peak signal-to-noise ratio, structure similarity index, feature similarity, and texture Haralick features were used to measure the performance difference between SP-CNN-T and SP-MRF-T algorithms in terms of the structure and tissue texture preservation, demonstrating the feasibility and the potential of the investigated ML approach.
Conclusions: Both training performance and image reconstruction results showed the feasibility of constructing CNN texture prior model and the potential of improving the structure preservation of the nodule comparing to our previous regional tissue-specific MRF texture prior model.
Bayesian theory lies down a sound framework for ultralow-dose computed tomography (ULdCT) image reconstruction with two terms for modeling the data statistical property and incorporating a priori knowledge for the tobe- reconstructed image. This study investigates the feasibility of using machine learning strategy, particularly the convolutional neural network (CNN), to construct a tissue-specific texture prior from previous full-dose CT (FdCT) and integrates the prior with the pre-log shift Poisson (SP) data property for Bayesian reconstruction of ULdCT images. The Bayesian reconstruction was implemented by an algorithm, called SP-CNN-T, and compared with our previous Markov random field (MRF) based tissue-specific texture prior algorithm, called SP-MRF-T. Both training performance and image reconstruction results showed the feasibility of constructing CNN texture prior model and the potential of improving the structure preservation of the nodule comparing to our previous regional tissue-specific MRF texture prior model. Quantitative structure similarity index (SSIM) and texture Haralick features (HF) were used to measure the performance difference between SP-CNN-T and SP-MRF-T algorithms, demonstrating the feasibility and the potential of the investigated machine learning approach.
Thin slice thickness reconstructions from helical Multi Detector-row CT (MDCT) scanning may suffer from windmill artifacts because of the under-sampling of the data in the z- or detector-row direction (which is essentially a Nyquist sampling issue). There are two strategies for windmill artifacts reduction: one is focusing on the CT system hardware design such as flying focal spot (FFS), the other is committed to correction using algorithms. Recently, numerous algorithms have been proposed to address this issue. One method aims to recover high-resolution images from thick-slice low-resolution images which are without windmill artifacts. Another method is an image domain post-processing method which can suppress windmill artifacts by using prior information such as total variation (TV). However, both two methods blur sharp edges and are unable to recover fine details. In this work, a super-resolution (SR) reconstruction method is developed by combining low rank and TV regularization (LRTV) to improve the z-axis resolution of MDCT in the post processing step. Hence, the SR reconstruction is formulated as an optimization problem which is solved effectively via alternating direction method of multipliers (ADMM). Thereafter, combining the high-resolution image with original reconstructed image, which is affected by windmill artifacts, can obtain a more accurate image. We evaluated our algorithm on Anke 16-slice helical CT scanner. The results demonstrate that the proposed method can achieve better windmill artifacts removal performance than the competing methods and simultaneously preserve fine details.
The X-ray computer tomography (CT) scanner has been extensively used in medical diagnosis. How to reduce radiation dose exposure while maintain high image reconstruction quality has become a major concern in the CT field. In this paper, we propose a statistical iterative reconstruction framework based on structure tensor total variation regularization for low dose CT imaging. An accelerated proximal forward-backward splitting (APFBS) algorithm is developed to optimize the associated cost function. The experiments on two physical phantoms demonstrate that our proposed algorithm outperforms other existing algorithms such as statistical iterative reconstruction with total variation regularizer and filtered back projection (FBP).
Despite the significantly practical utilities of interior tomography, it still suffers from severe degradation of direct current
(DC) shift artifact. Existing literature suggest to introducing prior information of object support (OS) constraint or the
zeroth order image moment, i.e., the DC value into interior reconstruction to suppress the shift artifact, while the prior
information is not always available in practice. Aimed at alleviating the artifacts without prior knowledge, in this paper,
we reported an approach on the estimation of the object support which could be employed to estimate the zeroth order
image moment, and hence facilitate the DC shift artifacts removal in interior reconstruction. Firstly, by assuming most of
the reconstructed object consists of soft tissues that are equivalent to water, we reconstructed a virtual OS that is
symmetrical about the interior region of interest (ROI) for the DC estimation. Hence the DC value can be estimated from
the virtual reconstruction. Secondly, a statistical iterative reconstruction incorporated with the sparse representation in
terms of learned dictionary and the constraint in terms of image DC value was adopted to solve the interior tomography.
Experimental results demonstrate that the relative errors of the estimated zeroth order image moment are 4.7% and 7.6%,
corresponding to the simulated data of a human thorax and the real data of a sheep lung, respectively. Reconstructed
images with the constraint of the estimated DC value exhibit greatly superior image quality to that
without DC value constraint.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.