Philip Stenner

Partial scan artifact reduction (PSAR) for the assessment of cardiac perfusion in dynamic phase‐correlated CT

Cardiac CT achieves its high temporal resolution by lowering the scan range from 2π to π plus fan.angle (partial scan). This, however, introduces CT-value variations depending on the start angle of the π range. These partial scan artifacts are in the order of a few HU and prevent the quantitative evaluation of perfusion measurements. Our PSAR algorithm corrects a dynamic phase-correlated scan without a priori information. In general a full scan does not suffer from partial scan artifacts since all projections in [0; 2π] contribute to the data. In order to maintain the optimum temporal resolution and the phase-correlation PSAR creates an artificial full scan p<inf>n</inf>^F by projection-wise averaging a set of neighboring partial scans p<inf>n</inf>^P from the same perfusion examination (typically N = 30 phase-correlated partial scans distributed over 20 s). Corresponding to the angular range of each partial scan we extract virtual partial scan data sets p<inf>n</inf>^V from the artificial full scan p<inf>n</inf>^F. A standard reconstruction yields the corresponding images f<inf>n</inf>^P, f<inf>n</inf>^F and f<inf>n</inf>^V. Subtracting the artificial full scan image f<inf>n</inf>^F from the virtual partial scan image f<inf>n</inf>^V yields an artifact image that can be used to correct the original partial scan image: f<inf>n</inf>^C = f<inf>n</inf>^P − (f<inf>n</inf>^V − f<inf>n</inf>^F). Our method has been validated with a simulated semi-anthropomorphic heart phantom and with clinical scans. For the simulated case real full scans have been performed to provide theoretical reference values. The improvement of the root mean square errors between the full scans and the partial, respectively corrected scans, is up to 54%. The phase-correlated data now appear accurate enough for a quantitative analysis of cardiac perfusion.

Dual energy exposure control (DEEC) for computed tomography: algorithm and simulation study.

DECT means acquiring the same object at two different energies, respectively two different tube voltages U1 and U2. The raw data q1 and q2 undergo a decomposition process of type p=p(q1,q2). The raw data p are reconstructed to obtain monochromatic images of the attenuation μ, of the object density ρ, or of a specific material distribution. Recent advances in DECT focus on noise reduction techniques [S. Richard and J. H. Siewerdsen, Med. Phys. (2), 586-600 (2008)] and enable high performance DECT such as lung nodule detection [Shkumat et al., Med. Phys. (2), 629-632 (2008)]. Given p and a raw data-based projection-wise patient dose estimation D(α) the authors determine the optimal tube current curves I1(α) and I2(α), with α being the view angle, which minimizes image noise for a given patient dose level. DEEC can perform online; I1(α) and I2(α) can be determined during the scan. Simulation studies using semianthropomorphic phantom data were carried out. In particular, functions p that generate μ-images and density images were evaluated. Image quality was compared to standard scans at U0=120kV (clinical CT) and U0=45kV (micro-CT) that were taken at the same dose level (D0=D1+D2) and identical spatial resolution. Appropriate choice of p(q1,q2) allows to obtain μ-images that show fewer artifacts and yield image noise levels comparable to the noise of the standard scan. The authors compared the standard scan to μ-images at 70keV, which is the effective energy used in clinical CT, and found optimal results with μ-images at 25keV for micro-CT. Nonoptimal choice of the decomposition function will, however, significantly increase image noise. In particular μ-images at 511keV, as needed for PET/CT attenuation correction, exhibit more than twice as much image noise as the standard scan. With DEEC, which guarantees best dose usage possible, monochromatic images are generated with only slightly increased noise levels at the same dose compared to a standard scan. The benefit of significantly decreased artifacts appears to allow using DEEC-generated monochromatic images in daily routine. Furthermore, DEEC is not restricted to DECT and the inherent tube current modulation algorithm may also be applied to single energy CT.

Rawdata-Based Dual Energy CT (DECT) from inconsistent scans

Dual energy CT (DECT) acquires an object at two different detected spectra w1(E) and w2(E). Rawdata-based techniques can be applied whenever the integration lines (rays) of both spectra are identical. Then, the rawdata q1 and q2 undergo a decomposition process of type p = p(q1, q2). For spiral cone-beam dual source CT (DSCT) and for micro-DSCT the integration lines are disjunct and therefore inconsistent. Hence, one typically uses image-based subtraction techniques that are inferior to the rawdata-based methods. Our technique provides consistent rawdata in two steps. Starting from standard reconstructions of both scans, polychromatic forward projections of the volumes 1 and 2 yield the desired consistent rawdata q1 and q2. The second step computes the decomposition p(q1, q2) and reconstructs these sinograms using filtered backprojection. All computations are done in parallel beam geometry. To avoid loosing spatial resolution the linear terms of p are handled in image domain. Only the higher order terms undergo the two-step process. The algorithm was run on a Cell Broadband Engine (CBE) at 3.2 GHz in a Playstation 3 (PS3) computer (Sony Computer Entertainment, Japan). To evaluate our approach we decompose patient rawdata acquired with a SOMATOM Definition clinical DSCT scanner (Siemens Medical Solutions, Forchheim, Germany). The PS3 processes 35 slices per second (5122 pixels, 512 views per half rotation, two forward projections, decomposition, filtering, one backprojection); a typical volume of thousand slices is processed in half a minute.

Dynamic iterative beam hardening correction (DIBHC) for an optimized assessment of cardiac perfusion in ECG-correlated CT

In cardiac perfusion examinations large concentrations of iodine in the ventricle cause beam hardening artifacts that lead to incorrect perfusion parameters. Beam hardening corrections are either implemented as simple precorrections which cannot account for higher order beam hardening effects, or as iterative approaches that are based on segmenting the original image into material distribution images. Conventional segmentation algorithms fail to clearly distinguish between iodine and bone. Our new algorithm, DIBHC, calculates the time-dependent iodine distribution by analyzing the voxel changes of a cardiac perfusion examination (typically N ¿ 30 ECG- correlated scans distributed over a total scan time T ¿ 20 s). These voxel dynamics are due to changes in contrast agent. This prior information allows to precisely distinguish between bone and iodine and is key to DIBHC where each iteration consists of a multi-material (soft tissue, bone, iodine) polychromatic forward projection, a raw data comparison and a filtered backprojection. Simulations with a semi-anthropomorphic dynamic phantom and clinical scans using a dual source CT scanner (2 × 128 slices, 100 kV, 160 mAs, 0.28 s) have been carried out. The uncorrected images suffer from beam hardening artifacts that appear as dark bands connecting large concentrations of iodine in the ventricle and bony structures. The CT-values of the affected tissue are typically underestimated by up to 20 HU. One iteration of DIBHC greatly reduces these artifacts yielding CT-value deviations of only 1 HU for the simulations and improvements of up to 56 HU for the measurements. DIBHC greatly reduces the beam hardening artifacts induced by the contrast agent dynamics (and those due to bone) now allowing for an improved calculation of perfusion parameters that are essential for quantifying myocardial perfusion.

Partial scan artifact reduction (PSAR) for the assessment of cardiac perfusion in dynamic phase-correlated CT

PURPOSE Cardiac CT achieves its high temporal resolution by lowering the scan range from 2pi to pi plus fan angle (partial scan). This, however, introduces CT-value variations, depending on the angular position of the pi range. These partial scan artifacts are of the order of a few HU and prevent the quantitative evaluation of perfusion measurements. The authors present the new algorithm partial scan artifact reduction (PSAR) that corrects a dynamic phase-correlated scan without a priori information. METHODS In general, a full scan does not suffer from partial scan artifacts since all projections in [0, 2pi] contribute to the data. To maintain the optimum temporal resolution and the phase correlation, PSAR creates an artificial full scan pn(AF) by projectionwise averaging a set of neighboring partial scans pn(P) from the same perfusion examination (typically N approximately 30 phase-correlated partial scans distributed over 20 s and n = 1, ..., N). Corresponding to the angular range of each partial scan, the authors extract virtual partial scans pn(V) from the artificial full scan pn(AF). A standard reconstruction yields the corresponding images fn(P), fn(AF), and fn(V). Subtracting the virtual partial scan image fn(V) from the artificial full scan image fn(AF) yields an artifact image that can be used to correct the original partial scan image: fn(C) = fn(P) - fn(V) + fn(AF), where fn(C) is the corrected image. RESULTS The authors evaluated the effects of scattered radiation on the partial scan artifacts using simulated and measured water phantoms and found a strong correlation. The PSAR algorithm has been validated with a simulated semianthropomorphic heart phantom and with measurements of a dynamic biological perfusion phantom. For the stationary phantoms, real full scans have been performed to provide theoretical reference values. The improvement in the root mean square errors between the full and the partial scans with respect to the errors between the full and the corrected scans is up to 54% for the simulations and 90% for the measurements. CONCLUSIONS The phase-correlated data now appear accurate enough for a quantitative analysis of cardiac perfusion.

Empirical Dual Energy Calibration (EDEC) for Cone-Beam Computed Tomography

Material-selective imaging using dual energy CT (DECT) heavily relies on well-calibrated material decomposition functions. These require the precise knowledge of the detected X-ray spectrum and even if this is exactly known the reliability of DECT will suffer from scattered radiation. We propose an empirical method to determine the proper decomposition function. In contrast to other decomposition algorithms our empirical dual energy calibration (EDEC) technique does neither require knowledge of the spectrum nor of the attenuation coefficients. The desired material-selective rawdata p1 and p2 are obtained as a function of the measured attenuation data q1 and q2 (one DECT scan = two rawdata sets) using a polynomial function whose coefficients are determined using a general least squares fit based on thresholded images of a calibration phantom. Assumptions on the calibration phantom size or of its positioning are not made. Once the decomposition coefficients are determined DECT rawdata can be decomposed by simply passing them through the polynomial. To demonstrate EDEC simulations of an oval CTDI phantom, a lung phantom and a thorax phantom were carried out and a physical phantom composed of water and calcium hydroxypatite was measured with a dedicated in vivo dual source micro-CT scanner (TomoScope 30s Duo, VAMP GmbH, Erlangen, Germany). The rawdata were decomposed into its components, reconstructed and the pixel values obtained were compared to the theoretical values. The determination of the calibration coefficients with EDEC is very robust and depends only slightly on the type of calibration phantom used. Images of the said test phantoms (simulations and measurements) show a nearly perfect agreement with the theoretical mu-values and density values. Since EDEC is an empirical technique it inherently compensates for scatter components, given that the calibration phantom is of similar size as the test objects. The empirical dual energy calibration technique is a pragmatic, simple and reliable calibration approach that produces highly quantitative DECT images.

Automatic exposure control (AEC) for dual energy computed tomography (DECT)

DECT means acquiring the same object at two different energies, respectively two different tube voltages U1 and U2. The rawdata q1 and q2 undergo a decomposition process of type p = p(q1, q2). The rawdata p are reconstructed to obtain monochromatic images of the attenuation mu, of the object density p or of a specific material distribution. Given p and a rawdata-based projection-wise patient dose estimation D (alpha) we determine the optimal tube current curves I1(alpha) and I2 (alpha), with alpha being the view angle, that minimizes image noise for a given patient dose level. AEC for DECT can perform online, I1(alpha) and I2 (alpha) can be determined during the scan. Simulation studies using semianthropomorphic phantom data were carried out. In particular functions p that generate mu-images and density images were evaluated. Image quality was compared to standard scans at Uo = 120 kV (clinical CT) and Uo = 45 kV (micro-CT) that were taken at the same dose level (Do = D1 + D2) and identical spatial resolution. Appropriate choice of p(q1,q2) allows to obtain mu images that are both artifact-free and show image noise levels comparable to the noise of the standard scan. Here, mu images at 25 keV (micro-CT) and 70 keV (clinical CT) turned out to be optimal. Non-optimal choice of the decomposition function will, however, significantly increase image noise. In particular mu images at 511 keV, as needed for PET/CT attenuation correction, exhibit more than twice as much image noise as the standard scan. With DECT-AEC, that guarantees best dose usage possible, monochromatic images are generated with only slightly increased noise levels at the same dose compared to a standard scan. The benefit of significantly decreased artifacts appears to allow using DECT-AEC-generated monochromatic images in daily routine.