Search

Understanding decisions of deep learning techniques is important. Especially in the medical field, the reasons for a decision in a classification task are as crucial as the pure classification results. In this article, we propose a new approach to compute relevant parts of a medical image. Knowing the relevant parts makes it easier to understand decisions. In our approach, a convolutional neural network is employed to learn structures of images of lung nodules. Then, an evolutionary algorithm is applied to compute a simplified version of an unknown image based on the learned structures by the convolutional neural network. In the simplified version, irrelevant parts are removed from the original image. In the results, we show simplified images which allow the observer to focus on the relevant parts. In these images, more than 50% of the pixels are simplified. The simplified pixels do not change the meaning of the images based on the learned structures by the convolutional neural network. An experimental analysis shows the potential of the approach. Besides the examples of simplified images, we analyze the run time development. Simplified images make it easier to focus on relevant parts and to find reasons for a decision. The combination of an evolutionary algorithm employing a learned convolutional neural network is well suited for the simplification task. From a research perspective, it is interesting which areas of the images are simplified and which parts are taken as relevant.

Erworben im Rahmen der Schweizer Nationallizenzen (http://www.nationallizenzen.ch) Purpose: Incidental aneurysms pose a challenge to physicians who need to decide whether or not to treat them. A statistical model could potentially support such treatment decisions. The aim of this study was to compare a previously developed aneurysm rupture logistic regression probability model (LRM) to other machine learning (ML) classifiers for discrimination of aneurysm rupture status. Methods: Hemodynamic, morphological, and patient-related information of 1631 cerebral aneurysms characterized by computational fluid dynamics simulations were used to train support vector machines (SVMs) with linear and RBF kernel (RBF-SVM), k-nearest neighbors (kNN), decision tree, random forest, and multilayer perceptron (MLP) neural network classifiers for predicting the aneurysm rupture status. The classifiers’ accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were evaluated and compared to the LRM using 249 test cases obtained from two external cohorts. Additionally, important variables were determined based on the random forest and weights of the linear SVM. Results: The AUCs of the MLP, LRM, linear SVM, RBF-SVM, kNN, decision tree, and random forest were 0.83, 0.82, 0.80, 0.81, 0.76, 0.70, and 0.79, respectively. The accuracy ranged between 0.76 (decision tree,) and 0.79 (linear SVM, RBF-SVM, and MLP). Important variables for predicting the aneurysm rupture status included aneurysm location, the mean surface curvature, and maximum flow velocity. Conclusion: The performance of the LRM was overall comparable to that of the other ML classifiers, confirming its potential for aneurysm rupture assessment. To further improve the predictions, additional information, e.g., related to the aneurysm wall, might be needed.

Many Deep Neural Networks (DNNs) are implemented with the single objective to achieve high classification scores. However, there can be additional objectives like the minimization of computational costs. This is especially important in the field of mobile computing where not only the computational power itself is a limiting factor but also each computation consumes energy affecting the battery life. Unfortunately, the determination of minimal structures is not straightforward.

International audience; Grid computing has been the subject of many large national and international IT projects. However, not all goals of these projects have been achieved. In particular, the number of users lags behind the initial forecasts laid out by proponents of grid technologies. This underachievement may have led to claims that the grid concept as a whole is on its way to being replaced by Cloud computing and various X-as-a-Service approaches. In this paper, we try to analyze the current situation and to identify promising directions for future grid development. Although there are shortcomings in current grid systems, we are convinced that the concept as a whole remains valid and can benefit from new developments, including Cloud computing. Furthermore, we strongly believe that some future applications will require the grid approach and that, as a result, further research is required in order to turn this concept into reliable, efficient and user-friendly computing platforms.

Big Data is a new label given to a diverse field of data intensive informatics in which the datasets are so large that they become hard to work with effectively. The term has been mainly used in two contexts, firstly as a technological challenge when dealing with dataintensive domains such as high energy physics, astronomy or internet search, and secondly as a sociological problem when data about us is collected and mined by companies such as Facebook, Google, mobile phone companies, retail chains and governments. In this paper we look at this second issue from a new perspective, namely how can the user gain awareness of the personally relevant part Big Data that is publicly available in the social web. The amount of user-generated media uploaded to the web is expanding rapidly and it is beyond the capabilities of any human to sift through it all to see which media impacts our privacy. Based on an analysis of social media in Flickr, Locr, Facebook and Google+, we discuss privacy implications and potential of the emerging trend of geo-tagged social media. We then present a concept with which users can stay informed about which parts of the social Big Data deluge is relevant to them.

A growing number of compute and data storage jobs is performed on remote resources. In a cloud environment the customer can't be sure where a particular job is physically executed and thus cannot rely on the security and confidentiality of the remote resource. A solution for this problem is operating on encrypted functions and encrypted data. This enables a customer to generate a program that can be executed by a third party, without revealing the underlying algorithm or the processed data. This helps securing applications and data in a distributed digital ecosystem. We present a method to compute a secret program on an untrusted resource using fully homomorphic encrypted circuits. We sketch an algebraic homomorphism as a cryptographic foundation and define a sample system architecture for which we provide a software implementation. Our concept solves the problems of encrypted storage access with encrypted addresses and encrypted branching: in contrast to other approaches, like static one-pass circuit simulations, our system supports dynamic parameters and non-linear programs, that render branch-decisions at runtime and cannot be represented in a circuit with hard-wired in-circuit parameters and data. Our implementation comprises the runtime environment for an encrypted program and an assembler to generate the encrypted machine code.

The development of business models is of great importance for the German Grid inititive D-Grid, which is trying to establish a sustainable and self-financing long-term operating e-science infrastructure all over Germany. To guarantee the continuity and future growth of this infrastructure it is necessary to become independent of limited public funding and achieve the required revenues by chargeable services, support and developments. The business model of the German Grid Support Organization comprises the development, provision and support of generic services for the usage by all Grid communities within D-Grid. The realization of these business models is based on the development of a comprehensive accounting service that enables a seamless recording of the usage of all resources. The special challenge lies in the heterogeneous middlewares, used within D-Grid. The paper describes the development of a highly integrative accounting system as a component of the service portfolio of the German Grid Support Organization.

Within the COMED project, medical workstations have been developed which provide computer-assisted methods for image management, 2D and 3D image visualization and processing, teleconferencing and data communication An "user-friendly" human-computer interaction has to be provided.