Thomas Netsch

SENSE-DTI at 3 T

While holding vast potential, diffusion tensor imaging (DTI) with single‐excitation protocols still faces serious challenges. Limited spatial resolution, susceptibility to magnetic field inhomogeneity, and low signal‐to‐noise ratio (SNR) may be considered the most prominent limitations. It is demonstrated that all of these shortcomings can be effectively mitigated by the transition to parallel imaging technology and high magnetic field strength. Using the sensitivity encoding (SENSE) technique at 3 T, brain DTI was performed in nine healthy volunteers. Despite enhanced field inhomogeneity, parallel acquisition permitted both controlling geometric distortions and enhancing spatial resolution up to 0.8 mm in‐plane. Heightened SNR requirements were met in part by high base sensitivity at 3 T. A further significant increase in SNR efficiency was accomplished by SENSE acquisition, exploiting enhanced encoding speed for echo time reduction. Based on the resulting image data, high‐resolution tensor mapping is demonstrated. Magn Reson Med 51:230–236, 2004.

Scale-space signatures for the detection of clustered microcalcifications in digital mammograms

A method is described for the automated detection of microcalcifications in digitized mammograms. The method is based on the Laplacian scale-space representation of the mammogram only. First, possible locations of microcalcifications are identified as local maxima in the filtered image on a range of scales. For each finding, the size and local contrast is estimated, based on the Laplacian response denoted as the scale-space signature. A finding is marked as a microcalcification if the estimated contrast is larger than a predefined threshold which depends on the size of the finding. It is shown that the signature has a characteristic peak, revealing the corresponding image features. This peak can be robustly determined. The basic method is significantly improved by consideration of the statistical variation of the estimated contrast, which is the result of the complex noise characteristic of the mammograms. The method is evaluated with the Nijmegen database and compared to other methods using these mammograms. Results are presented as the free-response receiver operating characteristic (FROG) performance. At a rate of one false positive cluster per image the method reaches a sensitivity of 0.84, which is comparable to the best results achieved so far.

Quantitative evaluation of image-based distortion correction in diffusion tensor imaging

A statistical method for the evaluation of image registration for a series of images based on the assessment of consistency properties of the registration results is proposed. Consistency is defined as the residual error of the composition of cyclic registrations. By combining the transformations of different algorithms the consistency error allows a quantitative comparison without the use of ground truth, specifically, it allows a determination as to whether the algorithms are compatible and hence provide comparable registrations. Consistency testing is applied to evaluate retrospective correction of eddy current-induced image distortion in diffusion tensor imaging of the brain. In the literature several image transformations and similarity measures have been proposed, generally showing a significant reduction of distortion in side-by-side comparison of parametric maps before and after registration. Transformations derived from imaging physics and a three-dimensional affine transformation as well as mutual information (MI) and local correlation (LC) similarity are compared to each other by means of consistency testing. The dedicated transformations could not demonstrate a significant difference for more than half of the series considered. LC similarity is well-suited for distortion correction providing more consistent registrations which are comparable to MI.

Automatic Image-Driven Segmentation of the Ventricles in Cardiac Cine MRI

To propose and to evaluate a novel method for the automatic segmentation of the hearts two ventricles from dynamic (“cine”) short‐axis “steady state free precession” (SSFP) MR images. This segmentation task is of significant clinical importance. Previously published automated methods have various disadvantages for routine clinical use.

Towards real-time multi-modality 3-D medical image registration

Intensity value-based registration is a widely used technique for the spatial alignment of medical images. Generally, the registration transformation is determined by iteratively optimizing a similarity measure calculated from the grey values of both images. However, such algorithms may have high computational costs, especially in the case of multi-modality registration, which makes their integration into systems difficult. At present, registration based on mutual information (MI) still requires computation times of the order of several minutes. In this contribution we focus on a new similarity measure based on local correlation (LC) which is well-suited for numerical optimization. We show that LC can be formulated as a least-squares criterion which allows the use of dedicated methods. Thus, it is possible to register MR neuro perfusion time-series (128/sup 2//spl times/30 voxel, 40 images) on a moderate workstation in real-time: the registration of an image takes about 500 ms and is therefore several times faster than image acquisition time. For the registration of CT-MR images (512/sup 2//spl times/87 CT 256/sup 2//spl times/128 MR) a multiresolution framework is used. On top of the decomposition, which requires 47 s of computation time, the optimization with an algorithm based on Ml previously described in the literature takes 97 s. In contrast, the proposed approach only takes 13 s, corresponding to a speedup about a factor of 7. Furthermore, we demonstrate that the superior computational performance of LC is not gained at the expense of accuracy. In particular experiments with dual contrast MR images providing ground truth for the registration show a comparable sub-voxel accuracy of LC and MI similarity.

B-spline registration of 3D images with Levenberg-Marquardt optimization

B-splines are a well-known approach for non-rigid image registration. Though successfully applied to various medical applications they exhibit a high computational complexity mainly because of the lack of dedicated optimization methods. In this work we focus on a Levenberg-Marquardt type optimization routine. As a similarity measure we use least-squares functionals such as the sum of squared differences, the cross-correlation and the local correlation measure, respectively. The latter is used for multi-modality registration tasks. The proposed registration algorithm consists of three main parts. In each iteration step one has to (a) build a linear system of equations, (b) solve this system and compute an update, (c) determine the step length for the following iteration step. Appropriate stopping criteria ensure the termination of the registration task. A standard approach for (c) and several modifications are investigated. Using a quadratic model we are able to avoid additional execution of (b) during the step length adaption. Several solvers (Cholesky, CG, pre-conditioning) for (b) have been evaluated. Also, modifications on the most time consuming task (a) are investigated, leading to a speed-up by a factor up to 30. Finally, the algorithm is embedded in a multi-scale framework (both on image and on parameter level) providing additional regularization, an increased capture range and speed-up. Convergence tests have been successfully applied for a priori known transformations. Feasibility of the proposed approach is also shown for clinical applications including PET-CT registrations (19 data sets) and MR mammography.

Gray-Value Based Registration of CT and MR Images by Maximization of Local Correlation

For gray-value based multi-modality registration the similarity measure is essential. Excellent results have been obtained with mutual information for various modality combinations. In this contribution we consider local correlation as similarity measure for multi-modality registration. Using a software phantom it is analyzed why local correlation is suitable for this registration task whereas direct gray-value correlation itself is usually not. It is shown that registration with local correlation can be done using only a fraction of the image volume offering an opportunity to accelerate the algorithm. Within validation, registration of the phantom images, two simultaneously acquired dual contrast MR images, and a clinical CT-MR data set has been studied. For comparison, the data sets have also been registered with mutual information. The results show that not only mutual information, but also local correlation is suitable for gray-value based multi-modality registration.

Estimation of Organ Motion from 4D CT for 4D Radiation Therapy Planning of Lung Cancer

The goal of this paper is to automatically estimate the motion of the tumor and the internal organs from 4D CT and to extract the organ surfaces. Motion induced by breathing and heart beating is an important uncertainty in conformal external beam radiotherapy (RT) of lung tumors. 4D RT aims at compensating the geometry changes during irradiation by incorporating the motion into the treatment plan using 4D CT imagery. We establish two different methods to propagate organ models through the image time series, one based on deformable surface meshes, and the other based on volumetric B-spline registration. The methods are quantitatively evaluated on 8 3D CT images of the full breathing cycle of a patient with manually segmented lungs and heart. Both methods achieve good overall results, with mean errors of 1.02–1.33 mm and 0.78–2.05 mm for deformable surfaces and B-splines respectively. The deformable mesh is fast (40 seconds vs. 50 minutes), but accommodation of the heart and the tumor is currently not possible. B-spline registration estimates the motion of all structures in the image and their interior, but is susceptible to motion artifacts in CT.

Correction of misaligned slices in multi-slice cardiovascular magnetic resonance using slice-to-volume registration

A popular technique to reduce respiratory motion for cardiovascular magnetic resonance is to perform a multi-slice acquisition in which a patient holds their breath multiple times during the scan. The feasibility of rigid slice-to-volume registration to correct for misalignments of slice stacks in such images due to differing breath-hold positions is explored. Experimental results indicate that slice-to-volume registration can compensate for the typical misalignments expected. Correction of slice misalignment results in anatomically more correct images, as well as improved left ventricular volume measurements. The interstudy reproducibility has also been improved reducing the number of samples needed for cardiac MR studies.

Correction of misaligned slices in multi-slice MR cardiac examinations by using slice-to-volume registration

One of the main challenges with magnetic resonance (MR) cardiac image acquisition is to account for cardiac motion due to respiration. A popular technique to reduce respiratory motion is to perform a multi-slice acquisition in which a patient holds their breath multiple times during the scan. This paper explores the feasibility of using rigid slice-to-volume registration to correct for misalignments of slice stacks in such images due to differing breath-hold positions. The experimental results indicate that slice-to-volume registration is sufficiently accurate and robust to compensate for the typical misalignments expected. We show that correction of misalignments in such data results in anatomically more correct images, as well as improved left ventricular volume measurements. It also shows that by correcting for misalignments in short axis (SA) images, one can improve the interstudy reproducibility and hence reduce the number of samples needed for cardiac MR studies to show the same statistical significance