A proper spatial characterization of a laser beam profile is indisputably important for any laser-mater experiment as well as for protection of beamline optical elements. Method of ablation and desorption imprints provides thorough beam profile analysis applicable to a broad range of photon energies. This method, however, often requires up to thousands of shots which must be then manually analyzed. Here we present method based on deep learning image segmentation model which is able to substitute human element currently indispensable in this time-consuming ex situ post processing. It is a part of AbloCAM project – an universal device for semi-automatic beam profile analysis.

We present a fast hierarchical method to detect a presence of cancerous tissue in histological images. The image is not examined in detail everywhere but only inside several small regions of interest, called glimpses. The final classification is done by aggregating classification scores from a CNN on leaf glimpses at the highest resolution. Unlike in existing attention-based methods, the glimpses form a tree structure, low resolution glimpses determining the location of several higher resolution glimpses using weighted sampling and a CNN approximation of the expected scores. We show that it is possible to perform the classification with just a small number of glimpses, leading to an important speedup with only a small performance deterioration. Learning is possible using image labels only, as in the multiple instance learning (MIL) setting.

We address the task of detecting cancer in histological slide images based on training with weak, slide- and patch-level annotations, which are considerably easier to obtain than pixel-level annotations. we use CNN based patch-level descriptors and formulate the image classification task as a generalized multiple instance learning (MIL) problem. The generalization consists of requiring a certain number of positive instances in positive bags, instead of just one as in standard MIL. The descriptors are learned on a small number of patch-level annotations, while the MIL layer uses only image-level patches for training. We evaluate multiple generalized MIL methods on the H&E stained images of lymphatic nodes from the CAMELYON dataset and show that generalized MIL methods improve the classification results and outperform no-MIL methods in terms of slide-level AUC. Best classification results were achieved by the MI-SVM(k) classifier in combination with simple spatial Gaussian aggregation, achieving AUC 0.962. However, MIL did not outperform methods trained on pixel-level segmentations.

We address the task of automatic detection of lesions caused by multiple myeloma (MM) in femurs or other long bones from CT data. Such detection is already an important part of the multiple myeloma diagnosis and staging. However, it is so far performed mostly manually, which is very time consuming. We formulate the detection as a multiple instance learning (MIL) problem, where instances are grouped into bags and only bag labels are available. In our case, instances are regions in the image and bags correspond to images. This has the advantage of requiring only subject-level annotation (ground truth), which is much easier to get than voxel-level manual segmentation. We consider a generalization of the standard MIL formulation where we introduce a threshold on the number of required positive instances in positive bags. This corresponds better to the classification procedure used by the radiology experts and is more robust with respect to false positive instances. We extend several existing MIL algorithms to solve the generalized case by estimating the threshold during learning. We compare the proposed methods with the baseline method on a dataset of 220 subjects. We show that the generalized MIL formulation outperforms standard MIL methods for this task. For the task of distinguishing between healthy controls and MM patients with infiltrations, our best method makes almost no mistakes with a mean AUC of 0.982 and F1 = 0.965. We outperform the baseline method significantly in all conducted experiments.

Neurological soft signs (NSS) are core features of psychiatric disorders with significant neurodevelopmental origin. However, it is unclear whether NSS correlates are associated with neuropathological processes underlying the disease or if they are confounded by medication. Given that NSS are also present in healthy persons (HP), investigating HP could reveal NSS correlates, which are not biased by disease‐specific processes or drug treatment. Therefore, we used a combination of diffusion MRI analysis tools to provide a framework of specific white matter (WM) microstructure variations underlying NSS in HP.

Introduction: To evaluate the feasibility and accuracy of a semiautomatic, three-dimensional volume of interest (3D sphere) for measuring the apparent diffusion coefficient (ADC) in suspicious breast lesions compared to conventional single-slice two-dimensional regions of interest (2D ROIs).Method: This institutional-review-board-approved study included 56 participants with Breast Imaging Reporting and Data System 4/5 lesion. All received diffusion-weighted imaging magnetic resonance imaging prior to biopsy (6=0-1500 s/mm2). ADC values were measured in the lesions with both methods. Reproducibility and accuracies were compared.Results: Area under the curve was 0.93 [95% confidence interval (CI) 0.86-0.99] for the 3D sphere and 0.91 (95% CI 0.84-0.98) for the 2D ROIs without significantly differing reproducibility (P=.45).Conclusion: A semiautomatic 3D sphere could reliably estimate ADC values in suspicious breast lesions without significant difference compared to conventional 2D ROIs. (C) 2016 Elsevier Inc. All rights reserved.

Effective image-based artifact correction is an essential step in the analysis of diffusion MR images. Many current approaches are based on retrospective registration, which becomes challenging in the realm of high b-values and low signal-to-noise ratio, rendering the corresponding correction schemes more and more ineffective. We propose a novel registration scheme based on memetic search optimization that allows for simultaneous exploitation of different signal intensity relationships between the images, leading to more robust registration results. We demonstrate the increased robustness and efficacy of our method on simulated as well as in vivo datasets. In contrast to the state-of-art methods, the median target registration error (TRE) stayed below the voxel size even for high b-values (3000 s · mm -2 and higher) and low SNR conditions. We also demonstrate the increased precision in diffusion-derived quantities by evaluating Neurite Orientation Dispersion and Density Imaging (NODDI) derived measures on a in vivo dataset with severe motion artifacts. These promising results will potentially inspire further studies on metaheuristic optimization in diffusion MRI artifact correction and image registration in general.

Effective image-based correction of motion and other acquisition artifacts became an essential step in diffusion-weighted Magnetic Resonance Imaging (MRI) analysis as the micro-structural tissue analysis advances towards higher-order models. These come with increasing demands on the number of acquired images and the diffusion strength (b-value) yielding lower signal-to-noise ratios (SNR) and a higher susceptibility to artifacts. These conditions, however, render the current image-based correction schemes, which act retrospectively on the acquired images through pairwise registration, more and more ineffective. Following the hypothesis, that a more consequent exploitation of the different intensity relationships between the volumes would reduce registration outliers, a novel correction scheme based on memetic search is proposed. This scheme allows for incorporating all single image metrics into a multi-objective optimization approach. To allow a quantitative evaluation of registration precision, realistic synthetic data are constructed by extending a diffusion MRI simulation framework by motion and eddy-currents-caused artifacts. The increased robustness and efficacy of the multi-objective registration method is demonstrated on the synthetic as well as in-vivo datasets at different levels of motion and other acquisition artifacts. In contrast to the state-of-the-art methods, the average target registration error (TRE) remained below the single voxel size also at high b-values (3000 s.mm-2) and low signal-to-noise ratio in the moderately artifacted datasets. In the more severely artifacted data, the multi-objective method was able to eliminate most of the registration outliers of the state-of-the-art methods, yielding an average TRE below the double voxel size. In the in-vivo data, the increased precision manifested itself in the scalar measures as well as the fiber orientation derived from the higher-order Neurite Orientation Dispersion and Density Imaging (NODDI) model. For the neuronal fiber tracts reconstructed on the data after correction, the proposed method most closely resembled the ground-truth. The proposed multi-objective method has not only impact on the evaluation of higher-order diffusion models as well as fiber tractography and connectomics, but could also find application to challenging image registration problems in general.

Effective image-based artifact correction is an essential step in the application of higher order models in diffusion MRI. Most approaches rely on some kind of retrospective registration, which becomes increasingly challenging in the realm of high b-values and low signal-to-noise ratio (SNR), rendering standard correction schemes more and more ineffective. We propose a novel optimization scheme based on memetic search that allows for simultaneous exploitation of different signal intensity relationships between the images, leading to more robust registration results. We demonstrate the increased robustness and efficacy of our method on simulated as well as in-vivo datasets. The median TRE for an affine registration of b = 3000s/mm(2) acquisitions could be reduced from > 5 mm for a standard correction scheme to < 1 mm using our approach. In-vivo bootstrapping experiments revealed increased precision in all tensor-derived quantities.

Poster: ECR 2015 / C-0372 / Diffusion weighted imaging and dual-energy CT based assessment of treatment response after carbon ion therapy in hepatocellular carcinoma patients by: Simons1, T. Wilhelm1, D. Habermehl2, J. Hering1, M. Ganten1, H.-P. Schlemmer1; 1Heidelberg/DE, 2Munich/DE

In clinical settings, application of the most recent modelling techniques is usually unfeasible due to the limited acquisition time. Localised acquisitions enclosing only the object of interest by reducing the field-of-view (FOV) counteract the time limitation but pose new challenges to the subsequent processing steps like motion correction. We use datasets from the Human Connectome Project (HCP) to simulate head motion distorted reduced FOV acquisitions and present an evaluation of head motion correction approaches: the commonly used affine registration onto an unweighted reference image guided by the mutual information (MI) metric and a model-based approach, which uses reference images computed from approximated tensor data to improve the performance of the MI metric. While the standard approach using the MI metric yields up to 15% outliers (error>5 mm) and a mean spatial error above 1.5 mm, the model-based approach reduces the number of outliers (1%) and the spatial error significantly (p<0.01). The behavior is also reflected by the visual analysis of the MI metric. The evaluation shows that the MI metric is of very limited use for reduced FOV data post-processing. The model-based approach has proven more suitable in this context.

PURPOSE:Diagnosis of autism spectrum disorders (ASD) is difficult, as symptoms vary greatly and are difficult to quantify objectively. Recent work has focused on the assessment of non-invasive diffusion tensor imaging-based biomarkers that reflect the microstructural characteristics of neuronal pathways in the brain. While tractography-based approaches typically analyze specific structures of interest, a graph-based large-scale network analysis of the connectome can yield comprehensive measures of larger-scale architectural patterns in the brain. Commonly applied global network indices, however, do not provide any specificity with respect to functional areas or anatomical structures. Aim of this work was to assess the concept of network centrality as a tool to perform locally specific analysis without disregarding the global network architecture and compare it to other popular network indices.METHODS:We create connectome networks from fiber tractographies and parcellations of the human brain and compute global network indices as well as local indices for Wernicke's Area, Broca's Area and the Motor Cortex. Our approach was evaluated on 18 children suffering from ASD and 18 typically developed controls using magnetic resonance imaging-based cortical parcellations in combination with diffusion tensor imaging tractography.RESULTS:We show that the network centrality of Wernicke's area is significantly (p<0.001) reduced in ASD, while the motor cortex, which was used as a control region, did not show significant alterations. This could reflect the reduced capacity for comprehension of language in ASD.CONCLUSIONS:The betweenness centrality could potentially be an important metric in the development of future diagnostic tools in the clinical context of ASD diagnosis. Our results further demonstrate the applicability of large-scale network analysis tools in the domain of region-specific analysis with a potential application in many different psychological disorders.

Die fehlenden Referenzdaten für in vivo diffusion-MR Aufnahmen verhindern eine quantitative Auswertung von Vorverarbeitungsschritten wie der Korrektur der Kopfbewegung. Eine Simulation der Bewegung durch das nachträgliche Hinzufügen von Headmotion-Effekten durch Transformation der einzelnen Bildvolumen ist nur mit Translation möglich, da das diffusion-gewichtete Signals von der Orientierung der Faser abhängt. Um komplexere Bewegungen zu simulieren, stellen wir eine Erzeugung von Daten mit Kopfbewegung vor, auf Basis von Fiberfox, einem Werkzeug für die Generierung von Software-Phantomdaten. Wir zeigen, dass die Rotationskomponente verglichen zur Translation einen signifikanten (p < 0.01) Einfluss auf den resultierenden target registration error (TRE) hat und bei der Evaluation von Korrekturansätzen nicht vernachlässigt werden darf. Für höhere Diffusionswichtungen (b > 2000 s/mm2) übersteigt der TRE auf den durch Rotation augmentierten Daten die Voxelgröße. Fehler von solcher Größe beeinflussen jede weitere Auswertung, insbesondere die immer stärker im Fokus stehende genaue Analyse von mikro-strukturellen Gewebecharakteristiken.

To obtain a better insight in tissue microstructures using diffusion MRI, a high resolution and dense sampling of q-space is required. In clinical settings, however, this can often not be achieved due to limited acquisition time. Reduced field-of-view (FOV) approaches counteract this limitation but may pose a challenge for the post-processing steps such as motion and artifact correction. We present an evaluation of the potential problems that arise with reduced FOV data during the standard post-processing. The acquisition with reduced FOV is extracted from a full FOV dataset. We select three different registration tools to perform the standard data post-processing pipeline. We first evaluate the spatial error and then measure its impact on the tensor reconstruction as well as on the derived fractional anisotropy (FA). With reduced FOV images, the multi-scale registration methods showed high sensitivity to parameter selection and produced up to 30 % outliers. With an optimized parameter set, all registration methods yielded spatial errors of 1 mm (±0.572). The spatial error resulted in a mean error of 0.03 (±0.013) in the estimated FA values, and was thus of the same magnitude as group differences as they are typically reported in DTI studies. Regions with large FA differences were located especially in the corpus callosum. The evaluation indicates that diffusion-weighted MR acquisitions with reduced FOV require careful selection of registration parameters and also cautious interpretation when quantifying derived indices.

Early diagnosis of autism spectrum disorders (ASD) is difficult, as symptoms vary greatly and are difficult to quantify objectively. Recent work has focused on the assessment of non-invasive diffusion tensor imaging based biomarkers of the disease that reflect the microstructural characteristics of neuronal pathways in the brain. One of the most common symptoms is reduced language development. We quantify this reduction using a graph-based large-scale network analysis of the connectome with a focus on the language related areas of the brain. Using a group of 18 children suffering from ASD and 18 typically developed controls we show that the reduced capacity for comprehension of language in ASD is reflected in the significantly (p < 0.001) reduced network centrality of Wernicke's area while the motor cortex, that was used as a control region, did not show any significant alterations. These results suggest Wernicke's area is less well integrated within the brain connectome in children suffering from ASD.

There is an increasing interest in connectomics as means to characterize the brain both in healthy controls and in disease. Connectomics strongly relies on graph theory to derive quantitative network related parameters from data. So far only a limited range of possible parameters have been explored in the literature. In this work, we utilize a broad range of global statistic measures combined with supervised machine learning and apply it to a group of 16 children with autism spectrum disorders (ASD) and 16 typically developed (TD) children, which have been matched for age, gender and IQ. We demonstrate that 86.7 % accuracy is achieved in distinguishing between ASD patients and the TD control using highly discriminative graph features in a supervised machine learning setting.

Die moderne medizinische Bildgebung ermöglicht immer detailliertere Daten, deren Weiterverarbeitung sich um so zeitaufwendiger gestaltet. Von der Medizin werden jedoch immer schnellere Bildverarbeitungsalgorithmen gefordert. Um diese Forderung zu erfüllen, müssen alle zur Verfügung stehenden Ressourcen genutzt werden. Die Grafikkarte ist eine dieser verfügbaren Ressourcen und kann für die Parallelisierung von Bildverarbeitungsalgorithmen herangezogen werden. Damit die Bildverarbeitungsprozesse unkompliziert auf die Graphikkarte ausgelagert und parallel berechnet werden können, wird in diesem Beitrag eine Hardware-unabhängige Erweiterung des Medical Imaging Interaction Toolkit vorgestellt. Die Ergebnisse zeigen eine wesentliche Beschleunigung der Algorithmen auf der Grafikkarte.

PURPOSE:Bronchoscopic interventions, such as transbronchial needle aspiration (TBNA), are commonly performed procedures to diagnose and stage lung cancer. However, due to the complex structure of the lung, one of the main challenges is to find the exact position to perform a biopsy and to actually hit the biopsy target (e.g., a lesion). Today, most interventions are accompanied by fluoroscopy to verify the position of the biopsy instrument, which means additional radiation exposure for the patient and the medical staff. Furthermore, the diagnostic yield of TBNA is particularly low for peripheral lesions.METHODS:To overcome these problems the authors developed an image-guided, electromagnetic navigation system for transbronchial interventions. The system provides real time positioning information for the bronchoscope and a transbronchial biopsy instrument with only one preoperatively acquired computed tomography image. A twofold respiratory motion compensation method based on a particle filtering approach allows for guidance through the entire respiratory cycle. In order to evaluate our system, 18 transbronchial interventions were performed in seven ventilated swine lungs using a thorax phantom.RESULTS:All tracked bronchoscope positions were corrected to the inside of the tracheobronchial tree and 80.2% matched the correct bronchus. During regular respiratory motion, the mean overall targeting error for bronchoscope tracking and TBNA needle tracking was with compensation on 10.4 ± 1.7 and 10.8 ± 3.0 mm, compared to 14.4 ± 1.9 and 13.3 ± 2.7 mm with compensation off. The mean fiducial registration error (FRE) was 4.2 ± 1.1 mm.CONCLUSIONS:The navigation system with the proposed respiratory motion compensation method allows for real time guidance during bronchoscopic interventions, and thus could increase the diagnostic yield of transbronchial biopsy.

Motivated by the challenging questions of todays sinologists we are developing an automated system for processing of ancient Chinese inscriptions (sutras). As these inscriptions are not accessible due to location or damage our input data are noisy images of paper showing the texture of stones together with the inscriptions transfered by charcoal or pencil. Due to the vast amount and large sizes of the images we adopted highly parallelized -- and therefore high-performance -- anisotropic filtering using standard computer hardware. Additionally characters are localized/segmented for further processing by Optical Character Recognition. Real results for ancient Chinese inscriptions, which are mimicking non-standardized handwritings are shown.