Andreas Degenhard

Validation of nonrigid image registration using finite-element methods: application to breast MR images

Presents a novel method for validation of nonrigid medical image registration. This method is based on the simulation of physically plausible, biomechanical tissue deformations using finite-element methods. Applying a range of displacements to finite-element models of different patient anatomies generates model solutions which simulate gold standard deformations. From these solutions, deformed images are generated with a range of deformations typical of those likely to occur in vivo. The registration accuracy with respect to the finite-element simulations is quantified by co-registering the deformed images with the original images and comparing the recovered voxel displacements with the biomechanically simulated ones. The functionality of the validation method is demonstrated for a previously described nonrigid image registration technique based on free-form deformations using B-splines and normalized mutual information as a voxel similarity measure, with an application to contrast-enhanced magnetic resonance mammography image pairs. The exemplar nonrigid registration technique is shown to be of subvoxel accuracy on average for this particular application. The validation method presented here is an important step toward more generic simulations of biomechanically plausible tissue deformations and quantification of tissue motion recovery using nonrigid image registration. It will provide a basis for improving and comparing different nonrigid registration techniques for a diversity of medical applications, such as intrasubject tissue deformation or motion correction in the brain, liver or heart.

Evaluation of radiological features for breast tumour classification in clinical screening with machine learning methods

OBJECTIVE In this work, methods utilizing supervised and unsupervised machine learning are applied to analyze radiologically derived morphological and calculated kinetic tumour features. The features are extracted from dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) time-course data. MATERIAL The DCE-MRI data of the female breast are obtained within the UK Multicenter Breast Screening Study. The group of patients imaged in this study is selected on the basis of an increased genetic risk for developing breast cancer. METHODS The k-means clustering and self-organizing maps (SOM) are applied to analyze the signal structure in terms of visualization. We employ k-nearest neighbor classifiers (k-nn), support vector machines (SVM) and decision trees (DT) to classify features using a computer aided diagnosis (CAD) approach. RESULTS Regarding the unsupervised techniques, clustering according to features indicating benign and malignant characteristics is observed to a limited extend. The supervised approaches classified the data with 74% accuracy (DT) and providing an area under the receiver-operator-characteristics (ROC) curve (AUC) of 0.88 (SVM). CONCLUSION It was found that contour and wash-out type (WOT) features determined by the radiologists lead to the best SVM classification results. Although a fast signal uptake in early time-point measurements is an important feature for malignant/benign classification of tumours, our results indicate that the wash-out characteristics might be considered as important.

Validation of Non-rigid Registration Using Finite Element Methods

We present a novel validation method for non-rigid registration using a simulation of deformations based on biomechanical modelling of tissue properties. This method is tested on a previously developed non-rigid registration method for dynamic contrast enhanced Magnetic Resonance (MR) mammography image pairs [1]. We have constructed finite element breast models and applied a range of displacements to them, with an emphasis on generating physically plausible deformations which may occur during normal patient scanning procedures. From the finite element method (FEM) solutions, we have generated a set of deformed contrast enhanced images against which we have registered the original dynamic image pairs. The registration results have been successfully validated at all breast tissue locations by comparing the recovered displacements with the biomechanical displacements. The validation method presented in this paper is an important tool to provide biomechanical gold standard deformations for registration error quantification, which may also form the basis to improve and compare different non-rigid registration techniques for a diversity of medical applications.

A method for linking computed image features to histological semantics in neuropathology

In medical image analysis the image content is often represented by features computed from the pixel matrix in order to support the development of improved clinical diagnosis systems. These features need to be interpreted and understood at a clinical level of understanding Many features are of abstract nature, as for instance features derived from a wavelet transform. The interpretation and analysis of such features are difficult. This lack of coincidence between computed features and their meaning for a user in a given situation is commonly referred to as the semantic gap. In this work, we propose a method for feature analysis and interpretation based on the simultaneous visualization of feature and image domain. Histopathological images of meningiomas WHO (World Health Organization) grade I are represented by features derived from color transforms and the Discrete Wavelet Transform. The wavelet-based feature space is then visualized and explored using unsupervised machine learning methods. We show how to analyze and select features according to their relevance for the description of clinically relevant characteristics.

A method for the comparison of biomechanical breast models

Biomechanical models of the breast are being developed for a wide range of applications including image alignment tasks to improve diagnosis and therapy monitoring, imaging related studies of the biomechanical properties of lesions, and image guided interventions. In this paper we present a method to evaluate the accuracy with which biomechanical breast models based on finite element methods (FEM) can predict the displacements of tissue within the breast. Our experimental data was obtained by compressing the breast of a volunteer in a controlled manner, and the acquisition of MR images of the breast before and after compression. Non-rigid registration of these two MR volumes together with interactive identification of corresponding landmarks provided an independent estimate of the displacements. In addition, the non-rigid registration provided estimates of the displacements of the surface points (skin points) of the breast. The accuracy of the FEM models was evaluated using all or a subset of these surface displacements as boundary conditions. The influence of pectoral muscle movement on the performance of the FEM models was also investigated. Our initial results indicate that the accurate setting of the boundary conditions is more important than the actual choice of elastic properties in these compression scenarios. With the complete boundary conditions, the displacements agreed to within 2.6 mm for all FEM models on average. Assuming no movement at the posterior or the medial side of the breast, the accuracy of the FEM models deteriorated to worse than 4.6 mm for all models.

Quantitative evaluation of free-form deformation registration for dynamic contrast-enhanced MR mammography.

In this paper, we present an evaluation study of a set of registration strategies for the alignment of sequences of 3D dynamic contrast-enhanced magnetic resonance breast images. The accuracy of the optimal registration strategies was determined on unseen data. The evaluation is based on the simulation of physically plausible breast deformations using finite element methods and on contrast-enhanced image pairs without visually detectable motion artifacts. The configuration of the finite element model was chosen according to its ability to predict in vivo breast deformations for two volunteers. We computed transformations for ten patients with 12 simulated deformations each. These deformations were applied to the postcontrast image to model patient motion occurring between pre- and postcontrast image acquisition. The original precontrast images were registered to the corresponding deformed postcontrast images. The performance of several registration configurations (rigid, affine, B-spline based nonrigid, single-resolution, multi-resolution, and volume-preserving) was optimized for five of the ten patients. The images were most accurately aligned with volume-preserving single-resolution nonrigid registration employing 40 or 20 mm control point spacing. When tested on the remaining five patients the optimal configurations reduced the average mean registration error from 1.40 to 0.45 mm for the whole breast tissue and from 1.20 to 0.32 mm for the enhancing lesion. These results were obtained on average within 26 (81) min for 40 (20) mm control point spacing. The visual appearance of the difference images from 30 patients was significantly improved after 20 mm volume-preserving single-resolution nonrigid registration in comparison to no registration or rigid registration. No substantial volume changes within the region of the enhancing lesions were introduced by this nonrigid registration.

Validation of Volume-Preserving Non-rigid Registration: Application to Contrast-Enhanced MR-Mammography

In this paper, we present a validation study for volume preserving non-rigid registration of 3D contrast-enhanced magnetic resonance mammograms. This study allows for the first time to assess the effectiveness of a volume preserving constraint to improve registration accuracy in this context. The validation is based on the simulation of physically plausible breast deformations with biomechanical breast models (BBMs) employing finite element methods. We constructed BBMs for four patients with four different deformation scenarios each. These deformations were applied to the post-contrast image to simulate patient motion occurring between pre- and post-contrast image acquisition. The original pre-contrast images were registered to the corresponding BBM-deformed post-contrast images. We assessed the accuracy of two optimisation schemes of a non-rigid registration algorithm. The first solely aims to improve the similarity of the images while the second includes the minimisation of volume changes as another objective. We observed reductions in residual registration error at every resolution when constraining the registration to preserve volume. Within the contrast enhancing lesion, the best results were obtained with a control point spacing of 20mm, resulting in target registration errors below 0.5mm on average. This study forms an important milestone in making the non-rigid registration framework applicable for clinical routine use.

Letters: ISOLLE: LLE with geodesic distance

We propose an extension of the algorithm for nonlinear dimensional reduction locally linear embedding (LLE) based on the usage of the geodesic distance (ISOLLE). In LLE, each data point is reconstructed from a linear combination of its n nearest neighbors, which are typically found using the Euclidean distance. We show that the search for the neighbors performed with respect to the geodesic distance can lead to a more accurate preservation of the data structure. This is confirmed by experiments on both real-world and synthetic data.

Comparison of biomechanical breast models: a case study

We present initial results from evaluating the accuracy with which biomechanical breast models based on finite element methods can predict the displacements of tissue within the breast. We investigate the influence of different tissue elasticity values, Poissons ratios, boundary conditions, finite element solvers and mesh resolutions on one data set. MR images were acquired before and after compressing a volunteers breast gently. These images were aligned using a 3D non-rigid registration algorithm. The boundary conditions were derived from the result of the non-rigid registration or by assuming no patient motion at the deep or medial side. Three linear and two non-linear elastic material models were tested. The accuracy of the BBMs was assessed by the Euclidean distance of twelve corresponding anatomical landmarks. Overall, none of the tested material models was obviously superior to another regarding the set of investigated values. A major average error increase was noted for partially inaccurate boundary conditions at high Poissons ratios due to introduced volume change. Maximal errors remained, however, high for low Poissons ratio due to the landmarks closeness to the inaccurate boundary conditions. The choice of finite element solver or mesh resolution had almost no effect on the performance outcome.

Comparison between radiological and artificial neural network diagnosis in clinical screening.

The imaging protocol of the UK multicentre magnetic resonance imaging study for screening in women at genetic risk of breast cancer aims to assist in detecting and diagnosing malignant breast lesions. In this paper, we evaluate a three-layer, feed-forward, backpropagation neural network as an artificial radiological classifier using receiver operating characteristic (ROC) curve analysis and compare the results with those obtained using a proposed radiological scoring system for the study which currently supplements the radiologists clinical opinion, in comparison with histological diagnosis. Based on the 76 symptomatic cases evaluated, descriptive features scored by radiologists showed considerable overlap between benign and malignant, although some features such as irregular contours and heterogeneous enhancement were more often associated with malignant pathology. In this preliminary evaluation, ROC analysis showed that the proposed scoring scheme did not perform well, indicating further refinement is required. When all 23 features were used in the neural network, its performance was poorer than that of the scoring scheme. When only ten features were used, limited to descriptors of enhancement characteristics, the neural network performed similar to the scoring scheme. This comparison shows that the neural network approach to clinical diagnosis has considerable potential and warrants further development.