Maik Stille

Augmented Likelihood Image Reconstruction

The presence of high-density objects remains an open problem in medical CT imaging. Data of projections passing through objects of high density, such as metal implants, are dominated by noise and are highly affected by beam hardening and scatter. Reconstructed images become less diagnostically conclusive because of pronounced artifacts that manifest as dark and bright streaks. A new reconstruction algorithm is proposed with the aim to reduce these artifacts by incorporating information about shape and known attenuation coefficients of a metal implant. Image reconstruction is considered as a variational optimization problem. The afore-mentioned prior knowledge is introduced in terms of equality constraints. An augmented Lagrangian approach is adapted in order to minimize the associated log-likelihood function for transmission CT. During iterations, temporally appearing artifacts are reduced with a bilateral filter and new projection values are calculated, which are used later on for the reconstruction. A detailed evaluation in cooperation with radiologists is performed on software and hardware phantoms, as well as on clinically relevant patient data of subjects with various metal implants. Results show that the proposed reconstruction algorithm is able to outperform contemporary metal artifact reduction methods such as normalized metal artifact reduction.

Influence of metal segmentation on the quality of metal artifact reduction methods

In computed tomography, star shape artifacts are introduced by metal objects, which are inside a patients body. The quality of the reconstructed image can be enhanced by applying a metal artifact reduction method. Unfortunately, a method that removes all such artifacts in order to make the images valuable for medical diagnosis remains to be found. In this study, the influence of metal segmentation is investigated. A thresholding technique, which is the state of the art in the field, is compared with a manual segmentation. Results indicate that a more accurate segmentation can lead to a preservation of important anatomical details, which are of high value for medical diagnosis.

The effects of metal artifact reduction on the retrieval of attenuation values

Abstract Background The quality of CT slices can be drastically reduced in the presence of high‐density objects such as metal implants within the patients’ body due to the occurrence of streaking artifacts. Consequently, a delineation of anatomical structures might not be possible, which strongly influences clinical examination. Purpose The aim of the study is to clinically evaluate the retrieval of attenuation values and structures by the recently proposed Augmented Likelihood Image Reconstruction (ALIR) and linear interpolation in the presence of metal artifacts. Material and Methods A commercially available phantom was equipped with two steel inserts. At a position between the metal rods, which shows severe streaking artifacts, different human tissue‐equivalent inserts are alternately mounted. Using a single‐source computer tomograph, raw data with and without metal rods are acquired for each insert. Images are reconstructed using the ALIR algorithm and a filtered back projection with and without linear interpolation. Mean and standard deviation are compared for a region of interest in the ALIR reconstructions, linear interpolation results, uncorrected images with metal rods, and the images without metal rods, which are used as a reference. Furthermore, the reconstructed shape of the inserts is analyzed by comparing different profiles of the image. Results The measured mean and standard deviation values show that for all tissue classes, the metal artifacts could be reduced using the ALIR algorithm and the linear interpolation. Furthermore, the HU values for the different classes could be retrieved with errors below the standard deviation in the reference image. An evaluation of the shape of the inserts shows that the reconstructed object fits the shape of the insert accurately after metal artifact correction. Moreover, the evaluation shows a drop in the standard deviation for the ALIR reconstructed images compared to the reference images while reducing artifacts and keeping the shape of the inserts, which indicates a noise reduction ability of the ALIR algorithm. Conclusion HU values, which are distorted by metal artifacts, can be retrieved accurately with the ALIR algorithm and the linear interpolation approach. After metal artifact correction, structures, which are not perceptible in the original images due to streaking artifacts, are reconstructed correctly within the image using the ALIR algorithm. Furthermore, the ALIR produced images with a reduced noise level compared to reference images and artifact images. Linear interpolation results in a distortion of the investigated shapes and features remaining streaking artifacts.

Improvement of dose calculation in radiation therapy due to metal artifact correction using the augmented likelihood image reconstruction

Abstract Background Metal artifacts caused by high‐density implants lead to incorrectly reconstructed Hounsfield units in computed tomography images. This can result in a loss of accuracy in dose calculation in radiation therapy. This study investigates the potential of the metal artifact reduction algorithms, Augmented Likelihood Image Reconstruction and linear interpolation, in improving dose calculation in the presence of metal artifacts. Materials and Methods In order to simulate a pelvis with a double‐sided total endoprosthesis, a polymethylmethacrylate phantom was equipped with two steel bars. Artifacts were reduced by applying the Augmented Likelihood Image Reconstruction, a linear interpolation, and a manual correction approach. Using the treatment planning system Eclipse™, identical planning target volumes for an idealized prostate as well as structures for bladder and rectum were defined in corrected and noncorrected images. Volumetric modulated arc therapy plans have been created with double arc rotations with and without avoidance sectors that mask out the prosthesis. The irradiation plans were analyzed for variations in the dose distribution and their homogeneity. Dosimetric measurements were performed using isocentric positioned ionization chambers. Results Irradiation plans based on images containing artifacts lead to a dose error in the isocenter of up to 8.4%. Corrections with the Augmented Likelihood Image Reconstruction reduce this dose error to 2.7%, corrections with linear interpolation to 3.2%, and manual artifact correction to 4.1%. When applying artifact correction, the dose homogeneity was slightly improved for all investigated methods. Furthermore, the calculated mean doses are higher for rectum and bladder if avoidance sectors are applied. Conclusion Streaking artifacts cause an imprecise dose calculation within irradiation plans. Using a metal artifact correction algorithm, the planning accuracy can be significantly improved. Best results were accomplished using the Augmented Likelihood Image Reconstruction algorithm.

Residual U-Net Convolutional Neural Network Architecture for Low-Dose CT Denoising

Abstract Low-dose CT has received increasing attention in the recent years and is considered a promising method to reduce the risk of cancer in patients. However, the reduction of the dosage leads to quantum noise in the raw data, which is carried on in the reconstructed images. Two different multilayer convolutional neural network (CNN) architectures for the denoising of CT images are investigated. ResFCN is based on a fully-convolutional network that consists of three blocks of 5×5 convolutions filters and a ResUNet that is trained with 10 convolutional blocks that are arranged in a multi-scale fashion. Both architectures feature a residual connection of the input image to ease learning. Training images are based on realistic simulations by using the XCAT phantom. The ResUNet approach shows the most promising results with a peak signal to noise ratio of 44.00 compared to ResFCN with 41.79.

Abstract: Retrieval of Attenuation Values by the Augmented Likelihood Image Reconstruction in the Presence of Metal Artefacts

Metal implants are able to cause severe artefacts in CT images due to physical effects such as scattering, total absorption, noise, or beamharding. Typically, the reconstructed images feature dark shadows around high-density objects as well as bright and dark streaks that may reduce the diagnostic value drastically. Within an extensive evaluation, the novel algorithm Augmented Likelihood Image Reconstruction has proven to reduce occurring artefacts accurately [1].

Metal artifact reduction by projection replacements and non-local prior image integration

Abstract The presence of high-density objects remains an open problem in medical CT imaging. Data of projections that passing through such objects are dominated by noise. Reconstructed images become less diagnostically conclusive because of pronounced artifacts that manifest as dark and bright streaks. A new reconstruction algorithm is proposed, which incorporates information gained from a prior image. Based on a non-local regularization, these information are used to reduce streaking artifacts. In an iterative scheme, the prior image is transformed in order to match intermediate results of the reconstruction by solving a registration problem. During iterations, temporally appearing artifacts are reduced with a bilateral filter and projection values passing through high-density objects are replaced by new calculated values, which are used further on for the reconstruction. Results show that the proposed algorithm significantly reduces streaking artifacts.