Marc Kachelrieß

Generalized multi-dimensional adaptive filtering for conventional and spiral single-slice, multi-slice, and cone-beam CT.

In modern computed tomography (CT) there is a strong desire to reduce patient dose and/or to improve image quality by increasing spatial resolution and decreasing image noise. These are conflicting demands since increasing resolution at a constant noise level or decreasing noise at a constant resolution level implies a higher demand on x-ray power and an increase of patient dose. X-ray tube power is limited due to technical reasons. We therefore developed a generalized multi-dimensional adaptive filtering approach that applies nonlinear filters in up to three dimensions in the raw data domain. This new method differs from approaches in the literature since our nonlinear filters are applied not only in the detector row direction but also in the view and in the z-direction. This true three-dimensional filtering improves the quantum statistics of a measured projection value proportional to the third power of the filter size. Resolution tradeoffs are shared among these three dimensions and thus are considerably smaller as compared to one-dimensional smoothing approaches. Patient data of spiral and sequential single- and multi-slice CT scans as well as simulated spiral cone-beam data were processed to evaluate these new approaches. Image quality was assessed by evaluation of difference images, by measuring the image noise and the noise reduction, and by calculating the image resolution using point spread functions. The use of generalized adaptive filters helps to reduce image noise or, alternatively, patient dose. Image noise structures, typically along the direction of the highest attenuation, are effectively reduced. Noise reduction values of typically 30%-60% can be achieved in noncylindrical body regions like the shoulder. The loss in image resolution remains below 5% for all cases. In addition, the new method has a great potential to reduce metal artifacts, e.g., in the hip region.

Normalized metal artifact reduction (NMAR) in computed tomography

PURPOSE While modern clinical CT scanners under normal circumstances produce high quality images, severe artifacts degrade the image quality and the diagnostic value if metal prostheses or other metal objects are present in the field of measurement. Standard methods for metal artifact reduction (MAR) replace those parts of the projection data that are affected by metal (the so-called metal trace or metal shadow) by interpolation. However, while sinogram interpolation methods efficiently remove metal artifacts, new artifacts are often introduced, as interpolation cannot completely recover the information from the metal trace. The purpose of this work is to introduce a generalized normalization technique for MAR, allowing for efficient reduction of metal artifacts while adding almost no new ones. The method presented is compared to a standard MAR method, as well as MAR using simple length normalization. METHODS In the first step, metal is segmented in the image domain by thresholding. A 3D forward projection identifies the metal trace in the original projections. Before interpolation, the projections are normalized based on a 3D forward projection of a prior image. This prior image is obtained, for example, by a multithreshold segmentation of the initial image. The original rawdata are divided by the projection data of the prior image and, after interpolation, denormalized again. Simulations and measurements are performed to compare normalized metal artifact reduction (NMAR) to standard MAR with linear interpolation and MAR based on simple length normalization. RESULTS Promising results for clinical spiral cone-beam data are presented in this work. Included are patients with hip prostheses, dental fillings, and spine fixation, which were scanned at pitch values ranging from 0.9 to 3.2. Image quality is improved considerably, particularly for metal implants within bone structures or in their proximity. The improvements are evaluated by comparing profiles through images and sinograms for the different methods and by inspecting ROIs. NMAR outperforms both other methods in all cases. It reduces metal artifacts to a minimum, even close to metal regions. Even for patients with dental fillings, which cause most severe artifacts, satisfactory results are obtained with NMAR. In contrast to other methods, NMAR prevents the usual blurring of structures close to metal implants if the metal artifacts are moderate. CONCLUSIONS NMAR clearly outperforms the other methods for both moderate and severe artifacts. The proposed method reliably reduces metal artifacts from simulated as well as from clinical CT data. Computationally efficient and inexpensive compared to iterative methods, NMAR can be used as an additional step in any conventional sinogram inpainting-based MAR method.

ECG-correlated image reconstruction from subsecond multi-slice spiral CT scans of the heart

Subsecond spiral computed tomography (CT) offers great potential for improving heart imaging. The new multi-row detector technology adds significantly to this potential. We therefore developed and validated dedicated cardiac reconstruction algorithms for imaging the heart with subsecond multi-slice spiral CT utilizing electrocardiogram (ECG) information. The single-slice cardiac z-interpolation algorithms 180 degrees CI and 180 degrees CD [Med. Phys. 25, 2417-2431 (1998)] were generalized to allow imaging of the heart for M-slice scanners. Two classes of algorithms were investigated: 180 degrees MCD (multi-slice cardio delta), a partial scan reconstruction of 180 degrees + delta data with a < phi (fan angle) resulting in effective scan times of 250 ms (central ray) when a 0.5 s rotation mode is available, and 180 degrees MCI (multi-slice cardio interpolation), a piecewise weighted interpolation between successive spiral data segments belonging to the same heart phase, potentially providing a relative temporal resolution of 12.5% of the heart cycle when a four-slice scanner is used and the table increment is chosen to be greater than or equal to the collimated slice thickness. Data segments are selected by correlation with the simultaneously recorded ECG signal. Theoretical studies, computer simulations, as well as patient measurements were carried out for a multi-slice scanner providing M = 4 slices to evaluate these new approaches and determine the optimal scan protocol. Both algorithms, 180 degrees MCD and 180 degrees MCI, provide significant improvements in image quality, including extremely arythmic cases. Artifacts in the reconstructed images as well as in 3D displays such as multiplanar reformations were largely reduced as compared to the standard z-interpolation algorithm 180 degrees MLI (multi-slice linear interpolation). Image quality appears adequate for precise calcium scoring and CT angiography of the coronary arteries with conventional subsecond multislice spiral CT. It turned out that for heart rates fH > or = 70 min(-1) the partial scan approach 180 degrees MCD yields unsatisfactory results as compared to 180 degrees MCI. Our theoretical considerations show that a freely selectable scanner rotation time chosen as a function of the patients heart rate, would further improve the relative temporal resolution and thus further reduce motion artifacts. In our case an additional 0.6 s mode besides the available 0.5 s mode would be very helpful. Moreover, if technically feasible, lower rotation times such as 0.3 s or even less would result in improved image quality. The use of multi-slice techniques for cardiac CT together with the new z-interpolation methods improves the quality of heart imaging significantly. The high temporal resolution of 180 degrees MCI is adequate for spatial and temporal tracking of anatomic structures of the heart (4D reconstruction).

Improved total variation-based CT image reconstruction applied to clinical data.

In computed tomography there are different situations where reconstruction has to be performed with limited raw data. In the past few years it has been shown that algorithms which are based on compressed sensing theory are able to handle incomplete datasets quite well. As a cost function these algorithms use the ℓ(1)-norm of the image after it has been transformed by a sparsifying transformation. This yields to an inequality-constrained convex optimization problem. Due to the large size of the optimization problem some heuristic optimization algorithms have been proposed in the past few years. The most popular way is optimizing the raw data and sparsity cost functions separately in an alternating manner. In this paper we will follow this strategy and present a new method to adapt these optimization steps. Compared to existing methods which perform similarly, the proposed method needs no a priori knowledge about the raw data consistency. It is ensured that the algorithm converges to the lowest possible value of the raw data cost function, while holding the sparsity constraint at a low value. This is achieved by transferring the step-size determination of both optimization procedures into the raw data domain, where they are adapted to each other. To evaluate the algorithm, we process measured clinical datasets. To cover a wide field of possible applications, we focus on the problems of angular undersampling, data lost due to metal implants, limited view angle tomography and interior tomography. In all cases the presented method reaches convergence within less than 25 iteration steps, while using a constant set of algorithm control parameters. The image artifacts caused by incomplete raw data are mostly removed without introducing new effects like staircasing. All scenarios are compared to an existing implementation of the ASD-POCS algorithm, which realizes the step-size adaption in a different way. Additional prior information as proposed by the PICCS algorithm can be incorporated easily into the optimization process.

Advanced single-slice rebinning in cone-beam spiral CT.

To achieve higher volume coverage at improved z-resolution in computed tomography(CT), systems with a large number of detector rows are demanded. However, handling an increased number of detector rows, as compared to today’s four-slice scanners, requires to accounting for the cone geometry of the beams. Many so-called cone-beam reconstruction algorithms have been proposed during the last decade. None met all the requirements of the medical spiral cone-beam CT in regard to the need for high image quality, low patient dose and low reconstruction times. We therefore propose an approximate cone-beam algorithm which uses virtual reconstruction planes tilted to optimally fit 180° spiral segments, i.e., the advanced single-slice rebinning (ASSR) algorithm. Our algorithm is a modification of the single-slice rebinning algorithm proposed by Noo et al. [Phys. Med. Biol. 44, 561–570 (1999)] since we use tilted reconstruction slices instead of transaxial slices to approximate the spiral path. Theoretical considerations as well as the reconstruction of simulated phantom data in comparison to the gold standard 180°LI (single-slice spiral CT) were carried out. Image artifacts, z-resolution as well as noise levels were evaluated for all simulated scanners. Even for a high number of detector rows the artifact level in the reconstructed images remains comparable to that of 180°LI. Multiplanar reformations of the Defrise phantom show none of the typical cone-beam artifacts usually appearing when going to larger cone angles. Imagenoise as well as the shape of the respective slice sensitivity profiles are equivalent to the single-slice spiral reconstruction,z-resolution is slightly decreased. The ASSR has the potential to become a practical tool for medical spiral cone-beam CT. Its computational complexity lies in the order of standard single-slice CT and it allows to use available 2D backprojection hardware.

Frequency split metal artifact reduction (FSMAR) in computed tomography.

PURPOSE The problem of metal artifact reduction (MAR) is almost as old as the clinical use of computed tomography itself. When metal implants are present in the field of measurement, severe artifacts degrade the image quality and the diagnostic value of CT images. Up to now, no generally accepted solution to this issue has been found. In this work, a method based on a new MAR concept is presented: frequency split metal artifact reduction (FSMAR). It ensures efficient reduction of metal artifacts at high image quality with enhanced preservation of details close to metal implants. METHODS FSMAR combines a raw data inpainting-based MAR method with an image-based frequency split approach. Many typical methods for metal artifact reduction are inpainting-based MAR methods and simply replace unreliable parts of the projection data, for example, by linear interpolation. Frequency split approaches were used in CT, for example, by combining two reconstruction methods in order to reduce cone-beam artifacts. FSMAR combines the high frequencies of an uncorrected image, where all available data were used for the reconstruction with the more reliable low frequencies of an image which was corrected with an inpainting-based MAR method. The algorithm is tested in combination with normalized metal artifact reduction (NMAR) and with a standard inpainting-based MAR approach. NMAR is a more sophisticated inpainting-based MAR method, which introduces less new artifacts which may result from interpolation errors. A quantitative evaluation was performed using the examples of a simulation of the XCAT phantom and a scan of a spine phantom. Further evaluation includes patients with different types of metal implants: hip prostheses, dental fillings, neurocoil, and spine fixation, which were scanned with a modern clinical dual source CT scanner. RESULTS FSMAR ensures sharp edges and a preservation of anatomical details which is in many cases better than after applying an inpainting-based MAR method only. In contrast to other MAR methods, FSMAR yields images without the usual blurring close to implants. CONCLUSIONS FSMAR should be used together with NMAR, a combination which ensures an accurate correction of both high and low frequencies. The algorithm is computationally inexpensive compared to iterative methods and methods with complex inpainting schemes. No parameters were chosen manually; it is ready for an application in clinical routine.

Technical advances in multi-slice spiral CT.

X-ray computerised tomography (CT) scanning with continuous patient transport has been established under the name Spiral CT since several years as the standard clinical examination procedure. This technique has been improved continuously with respect to scan speed, temporal response and z-axis resolution by the use of latest technical developments: Rotation times up to 0.5 s and multi-row detector array systems. Today detector systems with M + 4 simultaneously measured slices are available. We report about recent progress of spiral CT reconstruction algorithms that are based on multi-slice data. It is demonstrated that the new technology not only provides significant reduction in overall scan times and thereby of the CT scanner?s X-ray tube load; beyond that, the new technology allows CT imaging of the beating heart with high level image quality in standard clinical routine.

Kymogram detection and kymogram-correlated image reconstruction from subsecond spiral computed tomography scans of the heart

Subsecond single-slice, multi-slice or cone-beam spiral computed tomography (SSCT, MSCT, CBCT) offer great potential for improving heart imaging. Together with the newly developed phase-correlated cardiac reconstruction algorithms 180 degrees MCD and 180 degrees MCI [Med. Phys. 27, 1881-1902 (2000)] or related algorithms provided by the CT manufacturers, high image quality can be achieved. These algorithms require information about the cardiac motion, i.e., typically the simultaneously recorded electrocardiogram (ECG), to synchronize the reconstruction with the cardiac motion. Neither data acquired without ECG information (standard patients) nor acquisitions with corrupted ECG information can be handled adequately. We developed a method to extract the appropriate information about cardiac motion directly from the measured raw data (projection data). The so-called kymogram function is a measure of the cardiac motion as a function of time t or as a function of the projection angle alpha. In contrast to the ECG which is a global measure of the hearts electric excitation, the kymogram is a local measure of the heart motion at the z-position z(a) at projection angle a. The patients local heart rate as well as the necessary synchronization information to be used with phase-correlated algorithms can be extracted from the kymogram by using a series of signal processing steps. The kymogram information is shown to be adequate to substitute the ECG information. Computer simulations with simulated ECG and patient measurements with simultaneously acquired ECG were carried out for a multislice scanner providing M = 4 slices to evaluate these new approaches. Both the ECG function and the kymogram function were used for reconstruction. Both were highly correlated regarding the periodicity information used for reconstruction. In 21 out of 25 consecutive cases the kymogram approach was equivalent to the ECG-correlated reconstruction; only minor differences in image quality between both methods were observed. For one patient the synchronization information detected by the ECG monitor turned out to be wrong; here, the kymogram constituted the only approach that provided useful reconstructions. Patient studies with 12 and 16 slices indicate the usefulness of our approach for cone-beam CT scans. Kymogram-correlated reconstructions also appear to have the potential to improve imaging of pericardial lung areas in general.

Empirical cupping correction: A first-order raw data precorrection for cone-beam computed tomography

We propose an empirical cupping correction (ECC) algorithm to correct for CT cupping artifacts that are induced by nonlinearities in the projection data. The method is raw data based, empirical, and requires neither knowledge of the x-ray spectrum nor of the attenuation coefficients. It aims at linearizing the attenuation data using a precorrection function of polynomial form. The coefficients of the polynomial are determined once using a calibration scan of a homogeneous phantom. Computing the coefficients is done in image domain by fitting a series of basis images to a template image. The template image is obtained directly from the uncorrected phantom image and no assumptions on the phantom size or of its positioning are made. Raw data are precorrected by passing them through the once-determined polynomial. As an example we demonstrate how ECC can be used to perform water precorrection for an in vivo micro-CT scanner (TomoScope 30 s, VAMP GmbH, Erlangen, Germany). For this particular case, practical considerations regarding the definition of the template image are given. ECC strives to remove the cupping artifacts and to obtain well-calibrated CT values. Although ECC is a first-order correction and cannot compete with iterative higher-order beam hardening or scatter correction algorithms, our in vivo mouse images show a significant reduction of bone-induced artifacts as well. A combination of ECC with analytical techniques yielding a hybrid cupping correction method is possible and allows for channel-dependent correction functions.

Hyperfast parallel‐beam and cone‐beam backprojection using the cell general purpose hardware

Tomographic image reconstruction, such as the reconstruction of computed tomography projection values, of tomosynthesis data, positron emission tomography or SPECT events, and of magnetic resonance imaging data is computationally very demanding. One of the most time-consuming steps is the backprojection. Recently, a novel general purpose architecture optimized for distributed computing became available: the cell broadband engine (CBE). To maximize image reconstruction speed we modified our parallel-beam backprojection algorithm [two dimensional (2D)] and our perspective backprojection algorithm [three dimensional (3D), cone beam for flat-panel detectors] and optimized the code for the CBE. The algorithms are pixel or voxel driven, run with floating point accuracy and use linear (LI) or nearest neighbor (NN) interpolation between detector elements. For the parallel-beam case, 512 projections per half rotation, 1024 detector channels, and an image of size 512(2) was used. The cone-beam backprojection performance was assessed by backprojecting a full circle scan of 512 projections of size 1024(2) into a volume of size 512(3) voxels. The field of view was chosen to completely lie within the field of measurement and the pixel or voxel size was set to correspond to the detector element size projected to the center of rotation divided by square root of 2. Both the PC and the CBE were clocked at 3 GHz. For the parallel backprojection of 512 projections into a 512(2) image, a throughput of 11 fps (LI) and 15 fps (NN) was measured on the PC, whereas the CBE achieved 126 fps (LI) and 165 fps (NN), respectively. The cone-beam backprojection of 512 projections into the 512(3) volume took 3.2 min on the PC and is as fast as 13.6 s on the cell. Thereby, the cell greatly outperforms todays top-notch backprojections based on graphical processing units. Using both CBEs of our dual cell-based blade (Mercury Computer Systems) allows to 2D backproject 330 images/s and one can complete the 3D cone-beam backprojection in 6.8 s.