Thorsten Twellmann

An adaptive tissue characterization network for model-free visualization of dynamic contrast-enhanced magnetic resonance image data

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has become an important source of information to aid cancer diagnosis. Nevertheless, due to the multi-temporal nature of the three-dimensional volume data obtained from DCE-MRI, evaluation of the image data is a challenging task and tools are required to support the human expert. We investigate an approach for automatic localization and characterization of suspicious lesions in DCE-MRI data. It applies an artificial neural network (ANN) architecture which combines unsupervised and supervised techniques for voxel-by-voxel classification of temporal kinetic signals. The algorithm is easy to implement, allows for fast training and application even for huge data sets and can be directly used to augment the display of DCE-MRI data. To demonstrate that the system provides a reasonable assessment of kinetic signals, the outcome is compared with the results obtained from the model-based three-time-points (3TP) technique which represents a clinical standard protocol for analysing breast cancer lesions. The evaluation based on the DCE-MRI data of 12 cases indicates that, although the ANN is trained with imprecisely labeled data, the approach leads to an outcome conforming with 3TP without presupposing an explicit model of the underlying physiological process.

Human vs. machine: evaluation of fluorescence micrographs

To enable high-throughput screening of molecular phenotypes, multi-parameter fluorescence microscopy is applied. Object of our study is lymphocytes which invade human tissue. One important basis for our collaborative project is the development of methods for automatic and accurate evaluation of fluorescence micrographs. As a part of this, we focus on the question of how to measure the accuracy of microscope image interpretation, by human experts or a computer system. Following standard practice we use methods motivated by receiver operator characteristics to discuss the accuracies of human experts and of neural network-based algorithms. For images of good quality the algorithms achieve the accuracy of the medium-skilled experts. In images with increased noise, the classifiers are outperformed by some of the experts. Furthermore, the neural network-based cell detection is much faster than the human experts.

Image fusion for dynamic contrast enhanced magnetic resonance imaging.

BackgroundMultivariate imaging techniques such as dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) have been shown to provide valuable information for medical diagnosis. Even though these techniques provide new information, integrating and evaluating the much wider range of information is a challenging task for the human observer. This task may be assisted with the use of image fusion algorithms.MethodsIn this paper, image fusion based on Kernel Principal Component Analysis (KPCA) is proposed for the first time. It is demonstrated that a priori knowledge about the data domain can be easily incorporated into the parametrisation of the KPCA, leading to task-oriented visualisations of the multivariate data. The results of the fusion process are compared with those of the well-known and established standard linear Principal Component Analysis (PCA) by means of temporal sequences of 3D MRI volumes from six patients who took part in a breast cancer screening study.ResultsThe PCA and KPCA algorithms are able to integrate information from a sequence of MRI volumes into informative gray value or colour images. By incorporating a priori knowledge, the fusion process can be automated and optimised in order to visualise suspicious lesions with high contrast to normal tissue.ConclusionOur machine learning based image fusion approach maps the full signal space of a temporal DCE-MRI sequence to a single meaningful visualisation with good tissue/lesion contrast and thus supports the radiologist during manual image evaluation.

Model-free visualization of suspicious lesions in breast MRI based on supervised and unsupervised learning

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has become an important tool in breast cancer diagnosis, but evaluation of multitemporal 3D image data holds new challenges for human observers. To aid the image analysis process, we apply supervised and unsupervised pattern recognition techniques for computing enhanced visualizations of suspicious lesions in breast MRI data. These techniques represent an important component of future sophisticated computer-aided diagnosis (CAD) systems and support the visual exploration of spatial and temporal features of DCE-MRI data stemming from patients with confirmed lesion diagnosis. By taking into account the heterogeneity of cancerous tissue, these techniques reveal signals with malignant, benign and normal kinetics. They also provide a regional subclassification of pathological breast tissue, which is the basis for pseudo-color presentations of the image data. Intelligent medical systems are expected to have substantial implications in healthcare politics by contributing to the diagnosis of indeterminate breast lesions by non-invasive imaging.

Detection of suspicious lesions in dynamic contrast enhanced MRI data

Dynamic contrast-enhanced magnet resonance imaging (DCE-MRI) has become an important source of information to aid breast cancer diagnosis. Nevertheless, next to the temporal sequence of 3D volume data from the DCE-MRI technique, the radiologist commonly adducts information from other modalities for his final diagnosis. Thus, the diagnosis process is time consuming and tools are required to support the human expert. We investigate an automatic approach that detects the location and delineates the extent of suspicious masses in multi-temporal DCE-MRI data sets. It applies the state-of-the-art support vector machine algorithm to the classification of the short-time series associated with each voxel. The ROC analysis shows an increased specificity in contrast to standard evaluations techniques.

Evaluation of multiparameter micrograph analysis with synthetical benchmark images

To analyze multiparametric images of biological systems in vivo, advanced image processing, visualization and data mining tools are under development. To evaluate the algorithms, fit the parameters and assess the algorithms accuracy, benchmark data sets supplied with a ground truth label are required. Because manual evaluation results, done by expensive experts are hard to achieve, new innovative strategies for evaluation are needed to allow the computer scientists to evaluate their algorithms. We propose an algorithm for the generation of synthetical multiparameter images of cell bodies. The algorithm generates a stack of intensity images, with different grey value characteristics from a reasonable set of parameters. Each image is composition of single synthesized grey value images. Tuning the parameters allow an individual incorporation of expert knowledge and supplies the developer with ground truth labels. We describe and motivate the algorithm and present three applications of evaluation tools to synthetical micrographs. The results show the approachs high relevance to application oriented image analysis in biomedicine.

Peak intensity prediction for pmf mass spectra using support vector regression

With the increasing amount of data nowadays produced in the field of proteomics, automated approaches for reliable protein identification are highly desirable. One widely-used approach are protein mass fingerprints (PMFs) that allow database searching for the unknown protein, based on a MALDI-TOF mass spectrum of its tryptic digest. Current approaches and software packages for interpreting PMFs do rarely make use of peak intensities in the measured spectrum, mostly due to the difficulty of predicting peak intensities in the simulated mass spectra. In this work, we address the problem of predicting peak intensities in MALDI-TOF mass spectra, and we use regression support vector machines (ν-SVR) for this purpose. We compare the impact of different preprocessing and normalization modes such as binning and balancing data sets on prediction accuracy. Our preliminary results indicate that we can predict peak intensities using ν-SVR even from very small data sets. It is reasonable to assume that peak intensity prediction can greatly improve automated peptide identification.

An Adaptive Extended Colour Scale for Comparison of Pseudo Colouring Techniques for DCE-MRI Data

Pseudo colouring techniques are frequently used for analysing multivariate image data such as dynamic contrast enhanced mr images. Dedicated features of the high dimensional signal are mapped to pseudo colour codes and are superimposed on the image data. Nevertheless, the examination and comparison of different mapping functions is difficult, because the variability of the multivariate signal and pseudo colour scale have to be adequately and simultaneously presented. In this paper, we propose a setup for examination of different pseudo colour scales based on self organising maps. Thereby, the data distribution of the high dimensional signal is represented by a structured set of signal prototypes. Application of the pseudo colouring techniques to these prototypes leads to an extended colour scale which simultaneously gives a comprehensive display of the signal space together with the colour codes.

Maximum Contrast Classifiers

Within the Bayesian setting of classification we present a method for classifier design based on constrained density modelling. The approach leads to maximization of some contrast function, which measures the discriminative power of the class-conditional densities used for classification. By an upper bound on the density contrast the sensitivity of the classifiers can be increased in regions with low density differences which are usually most important for discrimination. We introduce a parametrization of the contrast in terms of modified kernel density estimators with variable mixing weights. In practice the approach shows some favourable properties: first, for fixed hyperparameters, training of the resulting Maximum Contrast Classifier (MCC) is achieved by linear programming for optimization of the mixing weights. Second for a certain choice of the density contrast bound and the kernel bandwidth, the maximum contrast solutions lead to sparse representations of the classifiers with good generalization performance, similar to the maximum margin solutions of support vector machines. Third the method is readily furnished for the general multi-class problem since training proceeds in the same way as in the binary case.

Spectral clustering for data categorization based on self-organizing maps

The exploration and categorization of large and unannotated image collections is a challenging task in the field of image retrieval as well as in the generation of appearance based object representations. In this context the Self-Organizing Map (SOM) has shown to be an efficient and scalable tool for the analysis of image collections based on low level features. Next to commonly employed visualization methods, clustering techniques have been recently considered for the aggregation of SOM nodes into groups in order to facilitate category specific data exploration. In this paper, spectral clustering based on graph theoretic concepts is employed for SOM based data categorization. The results are compared with those from the Neural Gas algorithm and hierarchical agglomerative clustering. Using SOMs trained on an eigenspace representation of the Columbia Object Image Library 20 (COIL20), the correspondence of the cluster data to a semantic reference grouping is calculated. Based on the Adjusted Rand Index it is shown that independent from the number of selected clusters, spectral clustering achieves a significantly higher correspondence to the reference grouping than any of the other methods.