Joscha Maier

Robust primary modulation-based scatter estimation for cone-beam CT.

PURPOSE Scattered radiation is one of the major problems facing image quality in flat detector cone-beam computed tomography (CBCT). Previously, a new scatter estimation and correction method using primary beam modulation has been proposed. The original image processing technique used a frequency-domain-based analysis, which proved to be sensitive to the accuracy of the modulator pattern both spatially and in amplitude as well as to the frequency of the modulation pattern. In addition, it cannot account for penumbra effects that occur, for example, due to the finite focal spot size and the scatter estimate can be degraded by high-frequency components of the primary image. METHODS In this paper, the authors present a new way to estimate the scatter using primary modulation. It is less sensitive to modulator nonidealities and most importantly can handle arbitrary modulator shapes and changes in modulator attenuation. The main idea is that the scatter estimation can be expressed as an optimization problem, which yields a separation of the scatter and the primary image. The method is evaluated using simulated and experimental CBCT data. The scattering properties of the modulator itself are analyzed using a Monte Carlo simulation. RESULTS All reconstructions show strong improvements of image quality. To quantify the results, all images are compared to reference images (ideal simulations and collimated scans). CONCLUSIONS The proposed modulator-based scatter reduction algorithm may open the field of flat detector-based imaging to become a quantitative modality. This may have significant impact on C-arm imaging and on image-guided radiation therapy.

Monitoring internal organ motion with continuous wave radar in CT.

PURPOSE To avoid motion artifacts in medical imaging or to minimize the exposure of healthy tissues in radiation therapy, medical devices are often synchronized with the patients respiratory motion. Todays respiratory motion monitors require additional effort to prepare the patients, e.g., mounting a motion belt or placing an optical reflector on the patients breast. Furthermore, they are not able to measure internal organ motion without implanting markers. An interesting alternative to assess the patients organ motion is continuous wave radar. The aim of this work is to design, implement, and evaluate such a radar system focusing on application in CT. METHODS The authors designed a radar system operating in the 860 MHz band to monitor the patient motion. In the intended application of the radar system, the antennas are located close to the patients body inside the table of a CT system. One receive and four transmitting antennas are used to avoid the requirement of exact patient positioning. The radar waves propagate into the patients body and are reflected at tissue boundaries, for example at the borderline between muscle and adipose tissue, or at the boundaries of organs. At present, the authors focus on the detection of respiratory motion. The radar system consists of the hardware mentioned above as well as of dedicated signal processing software to extract the desired information from the radar signal. The system was evaluated using simulations and measurements. To simulate the radar system, a simulation model based on radar and wave field equations was designed and 4D respiratory-gated CT data sets were used as input. The simulated radar signals and the measured data were processed in the same way. The radar system hardware and the signal processing algorithms were tested with data from ten volunteers. As a reference, the respiratory motion signal was recorded using a breast belt simultaneously with the radar measurements. RESULTS Concerning the measurements of the test persons, there is a very good correlation (ρ = 0.917) between the respiratory motion phases received by the radar system and the external motion monitor. Our concept of using an array of transmitting antennas turned out to be widely insensitive to the positioning of the test persons. A time shift between the respiratory motion curves recorded with the radar system and the motion curves from the external respiratory monitor was observed which indicates a slight difference between internal organ motion and motion detected by the external respiratory monitor. The simulations were in good accordance with the measurements. CONCLUSIONS A continuous wave radar operating in the near field of the antennas can be used to determine the respiratory motion of humans accurately. In contrast to trigger systems used today, the radar system is able to measure motion inside the body. If such a monitor was routinely available in clinical CT, it would be possible optimizing the scan start with respect to the respiratory state of the patient. Breathing commands would potentially widely be avoided, and as far as uncooperative patients or children are concerned, less sedation might be necessary. Further applications of the radar system could be in radiation therapy or interventional imaging for instance.

An efficient computational approach to model statistical correlations in photon counting x‐ray detectors

PURPOSE To introduce and evaluate an increment matrix approach (IMA) describing the signal statistics of energy-selective photon counting detectors including spatial-spectral correlations between energy bins of neighboring detector pixels. The importance of the occurring correlations for image-based material decomposition is studied. METHODS An IMA describing the counter increase patterns in a photon counting detector is proposed. This IMA has the potential to decrease the number of required random numbers compared to Monte Carlo simulations by pursuing an approach based on convolutions. To validate and demonstrate the IMA, an approximate semirealistic detector model is provided, simulating a photon counting detector in a simplified manner, e.g., by neglecting count rate-dependent effects. In this way, the spatial-spectral correlations on the detector level are obtained and fed into the IMA. The importance of these correlations in reconstructed energy bin images and the corresponding detector performance in image-based material decomposition is evaluated using a statistically optimal decomposition algorithm. RESULTS The results of IMA together with the semirealistic detector model were compared to other models and measurements using the spectral response and the energy bin sensitivity, finding a good agreement. Correlations between the different reconstructed energy bin images could be observed, and turned out to be of weak nature. These correlations were found to be not relevant in image-based material decomposition. An even simpler simulation procedure based on the energy bin sensitivity was tested instead and yielded similar results for the image-based material decomposition task, as long as the fact that one incident photon can increase multiple counters across neighboring detector pixels is taken into account. CONCLUSIONS The IMA is computationally efficient as it required about 10(2) random numbers per ray incident on a detector pixel instead of an estimated 10(8) random numbers per ray as Monte Carlo approaches would need. The spatial-spectral correlations as described by IMA are not important for the studied image-based material decomposition task. Respecting the absolute photon counts and thus the multiple counter increases by a single x-ray photon, the same material decomposition performance could be obtained with a simpler detector description using the energy bin sensitivity.

X-ray spectrum estimation for accurate attenuation simulation

Purpose: To estimate detected x‐ray spectra from transmission measurements of known attenuators that allow to accurately simulate the transmission in unknown attenuators. Methods: Starting from the established spectrum estimation method using the truncated singular value decomposition (TSVD) we extended the algorithm by incorporating prior knowledge about the statistical nature of the transmission data and about high‐frequency spectral components like characteristic peaks. Thereby our proposed approach requires only minimal prior knowledge, namely the energy positions of characteristic peaks or k‐edges, which are typically well‐known. This ensures that the final spectrum is not biased towards a given prior spectrum which is often observed in other methods. The new approach, prior truncated singular value decomposition (PTSVD), is compared to TSVD as well as the expectation–maximization (EM) method in a simulation and a measurement study. The resulting spectra are evaluated according to their ability to reproduce transmission data of attenuators that have not been included into the estimation process. Results: In case of noiseless simulated data, the PTSVD approach outperforms the existing methods in both, estimating the shape of the spectrum as well as providing a spectrum that reproduces the transmission data. Not surprisingly for increasing noise the ability of PTSVD to estimate the spectral shape worsens while it still performs best in reproducing the transmission data. This finding is also confirmed in the measurement study. Conclusion: Our new approach allows to estimate detected x‐ray spectra that accurately reproduce both transmission measurements that have and have not been included into the estimation process. It is less prone to noise compared to the established TSVD method and potentially leads to smaller transmission errors compared to EM for accurate transmission data while being less biased towards the given prior information.

Assessment of dedicated low-dose cardiac micro-CT reconstruction algorithms using the left ventricular volume of small rodents as a performance measure

PURPOSE Phase-correlated microcomputed tomography (micro-CT) imaging plays an important role in the assessment of mouse models of cardiovascular diseases and the determination of functional parameters as the left ventricular volume. As the current gold standard, the phase-correlated Feldkamp reconstruction (PCF), shows poor performance in case of low dose scans, more sophisticated reconstruction algorithms have been proposed to enable low-dose imaging. In this study, the authors focus on the McKinnon-Bates (MKB) algorithm, the low dose phase-correlated (LDPC) reconstruction, and the high-dimensional total variation minimization reconstruction (HDTV) and investigate their potential to accurately determine the left ventricular volume at different dose levels from 50 to 500 mGy. The results were verified in phantom studies of a five-dimensional (5D) mathematical mouse phantom. METHODS Micro-CT data of eight mice, each administered with an x-ray dose of 500 mGy, were acquired, retrospectively gated for cardiac and respiratory motion and reconstructed using PCF, MKB, LDPC, and HDTV. Dose levels down to 50 mGy were simulated by using only a fraction of the projections. Contrast-to-noise ratio (CNR) was evaluated as a measure of image quality. Left ventricular volume was determined using different segmentation algorithms (Otsu, level sets, region growing). Forward projections of the 5D mouse phantom were performed to simulate a micro-CT scan. The simulated data were processed the same way as the real mouse data sets. RESULTS Compared to the conventional PCF reconstruction, the MKB, LDPC, and HDTV algorithm yield images of increased quality in terms of CNR. While the MKB reconstruction only provides small improvements, a significant increase of the CNR is observed in LDPC and HDTV reconstructions. The phantom studies demonstrate that left ventricular volumes can be determined accurately at 500 mGy. For lower dose levels which were simulated for real mouse data sets, the HDTV algorithm shows the best performance. At 50 mGy, the deviation from the reference obtained at 500 mGy were less than 4%. Also the LDPC algorithm provides reasonable results with deviation less than 10% at 50 mGy while PCF and MKB reconstruction show larger deviations even at higher dose levels. CONCLUSIONS LDPC and HDTV increase CNR and allow for quantitative evaluations even at dose levels as low as 50 mGy. The left ventricular volumes exemplarily illustrate that cardiac parameters can be accurately estimated at lowest dose levels if sophisticated algorithms are used. This allows to reduce dose by a factor of 10 compared to todays gold standard and opens new options for longitudinal studies of the heart.

Deep scatter estimation (DSE): feasibility of using a deep convolutional neural network for real-time x-ray scatter prediction in cone-beam CT

The contribution of scattered x-rays to the acquired projection data is a severe issue in cone-beam CT (CBCT). Due to the large cone angle, scatter-to-primary ratios may easily be in the order of 1. The corresponding artifacts which appear as cupping or dark streaks in the CT reconstruction may impair the diagnostic value of the CT examination. Therefore, appropriate scatter correction is essential. The gold standard is to use a Monte Carlo photon transport code to predict the distribution of scattered x-rays which can be subtracted from the measurement subsequently. However, long processing times of Monte Carlo simulations prohibit them to be used routinely. To enable fast and accurate scatter estimation we propose the deep scatter estimation (DSE). It uses a deep convolutional neural network which is trained to reproduce the output of Monte Carlo simulations using only the acquired projection data as input. Once the network is trained, DSE performs in real-time. In the present study we demonstrate the feasibility of DSE using simulations of CBCT head scans at different tube voltages. The performance is tested on data sets that significantly differ from the training data. Thereby, the scatter estimates deviate less than 2% from the Monte Carlo ground truth. A comparison to kernel-based scatter estimation techniques, as they are used today, clearly shows superior performance of DSE while being similar in terms of processing time.

Organ-specific context-sensitive CT image reconstruction and display

In this work, we present a novel method to combine mutually exclusive CT image properties that emerge from different reconstruction kernels and display settings into a single organ-specific image reconstruction and display. We propose a context-sensitive reconstruction that locally emphasizes desired image properties by exploiting prior anatomical knowledge. Furthermore, we introduce an organ-specific windowing and display method that aims at providing a superior image visualization. Using a coarse-to-fine hierarchical 3D fully convolutional network (3D U-Net), the CT data set is segmented and classified into different organs, e.g. the heart, vasculature, liver, kidney, spleen and lung, as well as into the tissue types bone, fat, soft tissue and vessels. Reconstruction and display parameters most suitable for the organ, tissue type, and clinical indication are chosen automatically from a predefined set of reconstruction parameters on a per-voxel basis. The approach is evaluated using patient data acquired with a dual source CT system. The final context-sensitive images simultaneously link the indication-specific advantages of different parameter settings and result in images joining tissue-related desired image properties. A comparison with conventionally reconstructed and displayed images reveals an improved spatial resolution in highly attenuating objects and air while maintaining a low noise level in soft tissue in the compound image. The images present significantly more information to the reader simultaneously and dealing with multiple volumes may no longer be necessary. The presented method is useful for the clinical workflow and bears the potential to increase the rate of incidental findings.

A Feasibility Study of Automatic Multi-Organ Segmentation Using Probabilistic Atlas

Thoracic and abdominal multi-organ segmentation has been a challenging problem due to the inter-subject variance of human thoraxes and abdomens as well as the complex 3D intra-subject variance among organs. In this paper, we present a preliminary method for automatically segmenting multiple organs using non-enhanced CT data. The method is based on a simple framework using generic tools and requires no organ-specific prior knowledge. Specifically, we constructed a grayscale CT volume along with a probabilistic atlas consisting of six thoracic and abdominal organs: lungs (left and right), liver, kidneys (left and right) and spleen. A non-rigid mapping between the grayscale CT volume and a new test volume provided the deformation information for mapping the probabilistic atlas to the test CT volume. The evaluation with the 20 VISCERAL non-enhanced CT dataset showed that the proposed method yielded an average Dice coefficient of over 95% for the lungs, over 90% for the liver, as well as around 80% and 70% for the spleen and the kidneys respectively

A photon counting detector model based on increment matrices to simulate statistically correct detector signals

We present a novel increment matrix concept to simulate the correlations in an energy–selective photon counting detector. Correlations between the energy bins of neighboring detector pixels are introduced by scattered and fluorescence photons, together with the broadening of the induced charge clouds as they travel towards the electrodes, leading to charge sharing. It is important to generate statistically correct detector signals for the different energy bins to be able to realistically assess the detector’s performance in various tasks, e.g. material decomposition. Our increment matrix concept describes the counter increases in neighboring pixels on a single event level. Advantages of our model are the fact that much less random numbers are required than simulating single photons and that the increment matrices together with their probabilities have to be generated only once and can be stored for later use. The different occurring increment matrix sets and the corresponding probabilities are simulated using an analytic model of the photon–matter–interactions based on the photoelectric effect and Compton scattering, and the charge cloud drift, featuring thermal diffusion and Coulomb expansion of the charge cloud. The results obtained with this model are evaluated in terms of the spectral response for different detector geometries and the resulting energy bin sensitivity. Comparisons to published measured data and a parameterized detector model show both a good qualitative and quantitative agreement. We also studied the resulting covariance of reconstructed energy bin images.

Technical Note: Intrinsic Rawdata-based CT Misalignment Correction without Redundant Data

PURPOSE CT image reconstruction requires accurate knowledge of the used geometry or image quality might be degraded by misalignment artifacts. To overcome this issue, an intrinsic method, that is, a method not requiring a dedicated calibration phantom, to perform a raw data-based misalignment correction for CT is proposed herein that does not require redundant data and hence is applicable to measurements with less than 180 ∘ plus fan-angle of data. METHODS The forward projection of a volume reconstructed from a misaligned geometry resembles the acquired raw data if no redundant data are used, that is, if less than 180 ∘ plus fan-angle are used for image reconstruction. Hence, geometric parameters cannot be deduced from such data by an optimization of the geometry-dependent raw data fidelity. We propose to use a nonlinear transform applied to the reconstructed volume to introduce inconsistencies in the raw data that can be employed to estimate geometric parameters using less than 180 ∘ plus fan-angle of data. The proposed method is evaluated using simulations of the FORBILD head phantom and using actual measurements of a contrast-enhanced scan of a mouse acquired using a micro-CT. RESULTS Noisy simulations and actual measurements demonstrate that the proposed method is capable of correcting for artifacts arising from a misaligned geometry without redundant data while ensuring raw data fidelity. CONCLUSIONS The proposed method extends intrinsic raw data-based misalignment correction methods to an angular range of 180 ∘ or less and is thus applicable to systems with a limited scan range.