VCBM 17: Eurographics Workshop on Visual Computing for Biology and Medicine

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/2631687

Browse

Now showing 1 - 20 of 23

Multi-fiber Estimation and Tractography for Diffusion MRI using mixture of Non-central Wishart Distributions
(The Eurographics Association, 2017) Shakya, Snehlata; Gu, Xuan; Batool, Nazre; Özarslan, Evren; Knutsson, Hans; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Multi-compartmental models are popular to resolve intra-voxel fiber heterogeneity. One such model is the mixture of central Wishart distributions. In this paper, we use our recently proposed model to estimate the orientations of crossing fibers within a voxel based on mixture of non-central Wishart distributions. We present a thorough comparison of the results from other fiber reconstruction methods with this model. The comparative study includes experiments on a range of separation angles between crossing fibers, with different noise levels, and on real human brain diffusion MRI data. Furthermore, we present multi-fiber visualization results using tractography. Results on synthetic and real data as well as tractography visualization highlight the superior performance of the model specifically for small and middle ranges of separation angles among crossing fibers.
Design Considerations for Immersive Analytics of Bird Movements Obtained by Miniaturised GPS Sensors
(The Eurographics Association, 2017) Nim, Hieu T.; Sommer, Björn; Klein, Karsten; Flack, Andrea; Safi, Kamran; Nagy, Máté; Fiedler, Wolfgang; Wikelski, Martin; Schreiber, Falk; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Recent advances in miniaturising sensor tags allow to obtain high-resolution bird trajectories, presenting an opportunity for immersive close-up observation of individual and group behaviour in mid-air. The combination of geographical, environmental, and movement data is well suited for investigation in immersive analytics environments. We explore the benefits and requirements of a wide range of such environments, and illustrate a multi-platform immersive analytics solution, based on a tiled 3D display wall and head-mounted displays (Google Cardboard, HTC Vive and Microsoft Hololens). Tailored to biologists studying bird movement data, the immersive environment provides a novel interactive mode to explore the geolocational time-series data. This paper aims to inform the 3D visualisation research community about design considerations obtained from a real world data set in different 3D immersive environments. This work also contributes to ongoing research efforts to promote better understanding of bird migration and the associated environmental factors at the planet-level scale, thereby capturing the public awareness of environmental issues.
Visual Analytics of Missing Data in Epidemiological Cohort Studies
(The Eurographics Association, 2017) Alemzadeh, Shiva; Niemann, Uli; Ittermann, Till; Völzke, Henry; Schneider, Daniel; Spiliopoulou, Myra; Preim, Bernhard; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
We introduce a visual analytics solution to analyze and treat missing values. Our solution is based on general approaches to handle missing values, but is fine-tuned to the problems in epidemiological cohort study data. The most severe missingness problem in these data is the considerable dropout rate in longitudinal studies that limits the power of statistical analysis and the validity of study findings. Our work is inspired by discussions with epidemiologists and tries to add visual components to their current statistics-based approaches. In this paper we provide a graphical user interface for exploration, imputation and checking the quality of imputations.
Watergate: Visual Exploration of Water Trajectories in Protein Dynamics
(The Eurographics Association, 2017) Vad, Viktor; Byška, Jan; Jurcík, Adam; Viola, Ivan; Gröller, Eduard; Hauser, Helwig; Marques, Sérgio M.; Damborský, Jiří; Kozlíková, Barbora; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
The function of proteins is tightly related to their interactions with other molecules. The study of such interactions often requires to track the molecules that enter or exit specific regions of the proteins. This is investigated with molecular dynamics simulations, producing the trajectories of thousands of water molecules during hundreds of thousands of time steps. To ease the exploration of such rich spatio-temporal data, we propose a novel workflow for the analysis and visualization of large sets of water-molecule trajectories. Our solution consists of a set of visualization techniques, which help biochemists to classify, cluster, and filter the trajectories and to explore the properties and behavior of selected subsets in detail. Initially, we use an interactive histogram and a time-line visualization to give an overview of all water trajectories and select the interesting ones for further investigation. Further, we depict clusters of trajectories in a novel 2D representation illustrating the flows of water molecules. These views are interactively linked with a 3D representation where we show individual paths, including their simplification, as well as extracted statistical information displayed by isosurfaces. The proposed solution has been designed in tight collaboration with experts to support specific tasks in their scientific workflows. They also conducted several case studies to evaluate the usability and effectiveness of our new solution with respect to their research scenarios. These confirmed that our proposed solution helps in analyzing water trajectories and in extracting the essential information out of the large amount of input data.
Combining Pseudo Chroma Depth Enhancement and Parameter Mapping for Vascular Surface Models
(The Eurographics Association, 2017) Behrendt, Benjamin; Berg, Philipp; Preim, Bernhard; Saalfeld, Sylvia; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
The presence of depth cues in a visualization can be a great aid in understanding the structure and topology of a vessel tree. Pseudo Chromadepth is a well-known technique for enhancing depth perception in vascular 3D models. Since it strongly relies on the color channel to convey its depth cues, it is traditionally not suited for combined visualizations comprising color-encoded surface parameters. In this paper, we present and evaluate the use of a modified form of Pseudo Chromadepth that supports displaying additional surface parameters using the color channel while still increasing depth perception. This technique has been designed for the visualization of cerebral aneurysm models. We have combined a discretized color scale to visualize the surface parameter with the Pseudo Chromadepth color scale to convey depth using a Fresnel-inspired blending mask. To evaluate our approach, we have conducted two consecutive studies. The first was performed with 104 participants from the general public and the second with eleven experts in the fields of medical engineering and flow simulation. These studies show that Pseudo Chromadepth can be used in conjunction with color-encoded surface attributes to support depth perception as long as the color scale is chosen appropriately.
Comparative Visualization for Diffusion Tensor Imaging Group Study at Multiple Levels of Detail
(The Eurographics Association, 2017) Zhang, Changgong; Höllt, Thomas; Caan, Matthan W. A.; Eisemann, Elmar; Vilanova, Anna; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Diffusion Tensor Imaging (DTI) group studies often require the comparison of two groups of 3D diffusion tensor fields. The total number of datasets involved in the study and the multivariate nature of diffusion tensors together make this a challenging process. The traditional approach is to reduce the six-dimensional diffusion tensor to some scalar quantities, which can be analyzed with univariate statistical methods, and visualized with standard techniques such as slice views. However, this provides merely part of the whole story due to information reduction. If to take the full tensor information into account, only few methods are available, and they focus on the analysis of a single group, rather than the comparison of two groups. Simultaneously comparing two groups of diffusion tensor fields by simple juxtaposition or superposition is rather impractical. In this work, we extend previous work by Zhang et al. [ZCH 17] to visually compare two groups of diffusion tensor fields. To deal with the wealth of information, the comparison is carried out at multiple levels of detail. In the 3D spatial domain, we propose a detailson- demand glyph representation to support the visual comparison of the tensor ensemble summary information in a progressive manner. The spatial view guides analysts to select voxels of interest. Then at the detail level, the respective original tensor ensembles are compared in terms of tensor intrinsic properties, with special care taken to reduce visual clutter. We demonstrate the usefulness of our visual analysis system by comparing a control group and an HIV positive patient group.
Bone Fracture and Lesion Assessment using Shape-Adaptive Unfolding
(The Eurographics Association, 2017) Martinke, Hannes; Petry, Christian; Großkopf, Stefan; Suehling, Michael; Soza, Grzegorz; Preim, Bernhard; Mistelbauer, Gabriel; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
The assessment of rib bone fractures and lesions consists of many images that have to be thoroughly inspected slice-by-slice and rib-by-rib. Existing visualization methods, such as curved planar reformation (CPR), reduce the number of images to inspect and, in turn, the time spent per case. However, this task remains time-consuming and exhausting. In this paper, we propose a novel rib unfolding strategy that considers the cross-sectional shape of each rib individually and independently. This leads to shape-adaptive slices through the ribs. By aggregating these slices into a single image, we support radiologists with a concise overview visualization of the entire rib cage for fracture and lesion assessment. We present results of our approach along different cases of rib and spinal fractures as well as lesions. To assess the applicability of our method, we separately evaluated the segmentation (with 954 data sets) and the visualization (with two clinical coaches).
Exploration of Interventricular Septum Motion in Multi-Cycle Cardiac MRI
(The Eurographics Association, 2017) Tautz, Lennart; Hüllebrand, Markus; Steinmetz, Michael; Voit, Dirk; Frahm, Jens; Hennemuth, Anja; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Function of the heart, including interventricular septum motion, is influenced by respiration and contraction of the heart muscle. Recent real-time magnetic resonance imaging (MRI) can acquire multi-cycle cardiac data, which enables the analysis of the variation between heart cycles depending on factors such as physical stress or changes in respiration. There are no normal values for this variation in the literature, and there are no established tools for the analysis and exploration of such multi-cycle data available. We propose an analysis and exploration concept that automatically segments the left and right ventricle, extracts motion parameters and allows to interactively explore the results. We tested the concept using nine real-time MRI data sets, including one subject under increasing stress levels and one subject performing a breathing maneuver. All data sets could be automatically processed and then explored successfully, suggesting that our approach can robustly quantify and explore septum thickness in real-time MRI data.
Automatic Thrombus Detection in Non-enhanced Computed Tomography Images in Patients With Acute Ischemic Stroke
(The Eurographics Association, 2017) Löber, Patrick; Stimpel, Bernhard; Syben, Christopher; Maier, Andreas; Ditt, Hendrik; Schramm, Peter; Raczkowski, Boy; Kemmling, André; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
In case of an ischemic stroke, identifying and removing blood clots is crucial for a successful recovery. We present a novel method to automatically detect vascular occlusion in non-enhanced computed tomography (NECT) images. Possible hyperdense thrombus candidates are extracted by thresholding and connected component clustering. A set of different features is computed to describe the objects, and a Random Forest classifier is applied to predict them. Thrombus classification yields 98.7% sensitivity with 6.7 false positives per volume, and 91.1% sensitivity with 2.7 false positives per volume. The classifier assigns a clot probability > = 90% for every thrombus with a volume larger than 100 mm3 or with a length above 23 mm, and can be used as a reliable method to detect blood clots.
Visualizing and Exploring Dynamic Multichannel EEG Coherence Networks
(The Eurographics Association, 2017) Ji, Chengtao; Gronde, Jasper J. van de; Maurits, Natasha M.; Roerdink, Jos B. T. M.; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
An electroencephalography (EEG) coherence network represents functional brain connectivity, and is constructed by calculating the coherence between pairs of electrode signals as a function of frequency. Visualization of coherence networks can provide insight into unexpected patterns of cognitive processing and help neuroscientists to understand brain mechanisms. However, visualizing dynamic EEG coherence networks is a challenge for the analysis of brain connectivity, especially when the spatial structure of the network needs to be taken into account. In this paper, we present a design and implementation of a visualization framework for such dynamic networks. First, requirements for supporting typical tasks in the context of dynamic functional connectivity network analysis were collected from neuroscience researchers. In our design, we consider groups of network nodes and their corresponding spatial location for visualizing the evolution of the dynamic coherence network. We introduce an augmented timeline-based representation to provide an overview of the evolution of functional units (FUs) and their spatial location over time. This representation can help the viewer to identify relations between functional connectivity and brain regions, as well as to identify persistent or transient functional connectivity patterns across the whole timewindow. In addition, we modified the FU map representation to facilitate comparison of the behavior of nodes between consecutive FU maps. Our implementation also supports interactive exploration. The usefulness of our visualization design was evaluated by an informal user study. The feedback we received shows that our design supports exploratory analysis tasks well. The method can serve as an preprocessing step before a complete analysis of dynamic EEG coherence networks.
A Web-Based Tool for Cardiac Dyssynchrony Assessment on Ultrasound Data
(The Eurographics Association, 2017) Pezzatini, Daniele; Yagüe, Carlos; Rudenick, Paula; Blat, Josep; Bijnens, Bart; Camara, Oscar; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Cardiac resynchronization therapy (CRT) is a broadly used therapy in patients that suffers from heart failure (HF). The positive outcome of CRT depends strongly on the parameters criteria used to select patients and a lot of research has been done to introduce new and more reliable parameters. In this paper we propose an interactive tool to perform visual assessment and measurements on cardiac ultrasound images of patient with cardiac dyssynchrony. The tool is developed as a web application, allowing doctors to remotely access images and measurements.
Protein Tunnel Reprojection for Physico-Chemical Property Analysis
(The Eurographics Association, 2017) Malzahn, Jan; Kozlíková, Barbora; Ropinski, Timo; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Cavities are crucial for interactions of proteins with other molecules. While a variety of different cavity types exists, tunnels in particular play an important role, as they enable a ligand to deeply enter the active site of a protein where chemical reactions can undergo. Consequently, domain scientists are interested in understanding properties relevant for binding interactions inside molecular tunnels. Unfortunately, when inspecting a 3D representation of the molecule under investigation, tunnels are difficult to analyze due to occlusion issues. Therefore, within this paper we propose a novel reprojection technique that transforms the 3D structure of a molecule to obtain a 2D representation of the tunnel interior. The reprojection has been designed with respect to application-oriented design guidelines, we have identified together with our domain partners. To comply with these guidelines, the transformation preserves individual residues, while the result is capable of showing binding properties inside the tunnel without suffering from occlusions. Thus the reprojected tunnel interior can be used to display physico-chemical properties, e.g., hydrophobicity or amino acid orientation, of residues near a tunnel's surface. As these properties are essential for the interaction between protein and ligand, they can thus hint angles of attack for protein engineers. To demonstrate the benefits of the developed visualization, the obtained results are discussed with respect to domain expert feedback.
UI-Net: Interactive Artificial Neural Networks for Iterative Image Segmentation Based on a User Model
(The Eurographics Association, 2017) Amrehn, Mario; Gaube, Sven; Unberath, Mathias; Schebesch, Frank; Horz, Tim; Strumia, Maddalena; Steidl, Stefan; Kowarschik, Markus; Maier, Andreas; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
For complex segmentation tasks, fully automatic systems are inherently limited in their achievable accuracy for extracting relevant objects. Especially in cases where only few data sets need to be processed for a highly accurate result, semi-automatic segmentation techniques exhibit a clear benefit for the user. One area of application is medical image processing during an intervention for a single patient.We propose a learning-based cooperative segmentation approach which includes the computing entity as well as the user into the task. Our system builds upon a state-of-the-art fully convolutional artificial neural network (FCN) as well as an active user model for training. During the segmentation process, a user of the trained system can iteratively add additional hints in form of pictorial scribbles as seed points into the FCN system to achieve an interactive and precise segmentation result. The segmentation quality of interactive FCNs is evaluated. Iterative FCN approaches can yield superior results compared to networks without the user input channel component, due to a consistent improvement in segmentation quality after each interaction.
MRI Hip Joint Segmentation: A Locally Bhattacharyya Weighted Hybrid 3D Level Set Approach
(The Eurographics Association, 2017) Pham, Duc Duy; Morariu, Cosmin Adrian; Terheiden, Tobias; Landgraeber, Stefan; Jäger, Marcus; Pauli, Josef; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
In this paper, we propose a novel hybrid level set approach that locally balances the combined use of both Gradient Vector Flow and region based energy cost function by means of the Bhattacharyya coefficient. The local neighborhood of each contour point is naturally divided into an area encapsulated and one excluded by the contour. We propose utilizing the Bhattacharyya coefficient of the intensity distributions of these local areas to determine a point-wise weighting scheme for the curve propagation. The performance of our method regarding segmentation quality is evaluated on the segmentation of the hip joint in 10 MRI data sets. Our proposed method shows a clear improvement compared to conventional 3D level set approaches.
CT-Based Navigation Guidance for Liver Tumor Ablation
(The Eurographics Association, 2017) Alpers, Julian; Hansen, Christian; Ringe, Kristina; Rieder, Christian; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Image-guided thermal ablation procedures such as microwave ablation (MWA) or radiofrequency ablation (RFA) have become clinically accepted treatment options for liver tumors. The goal of these minimally invasive procedures is the destruction of focal liver malignancies using mostly needle-shaped instruments. Computed tomography (CT) imaging may be used to navigate the applicator to the target position in order to achieve complete tumor ablation. Due to limited image quality and resolution, the treatment target and risk structures may be hardly visible in intra-interventional CT-images, hampering verification of the intended applicator position. In this work, we propose a navigation guidance method based only on CT images to support the physician with additional information to reach the target position. Therefore, planning information extracted from pre-interventional images is fused with the current intra-interventional image. The visible applicator is extracted semi-automatically from the intra-interventional image. The localization of the needle instrument is used to guide the physician by display of the pathway, projection of anatomical structures, and correction suggestions. In an evaluation, we demonstrate the potential of the proposed method to improve the clinical success rate of complex liver tumor ablations while increasing the accuracy and reducing the number of intra-interventional CT images needed.
Application of Image Processing Functions for Brain Tumor Enhancement in Intraoperative Ultrasound Image Data
(The Eurographics Association, 2017) Chalopin, Claire; Mbuyamba, Elisee Ilunga; Aragon, Jesus Guillermo Cabal; Rodriguez, Juan Carlos Camacho; Arlt, Felix; Cervantes, Juan Gabriel Avina; Meixensberger, Juergen; Lindner, Dirk; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Intraoperative ultrasound (iUS) imaging supports neurosurgeons significantly during brain tumor operations. At the beginning of the intervention the integration of the iUS image data within the navigation system guides the surgeon by optimally planning the position and size of the skull opening. After tumor resection, the visualization of the iUS image data enables to identify possible tumor residuals. However, the iUS image data can be complex to interpret. Existing segmentation and registration functions were assembled into pipeline to enhance brain tumor contours in the 3D iUS image data. A brain tumor model, semi-automatically segmented in the preoperative MR data of patients, is rigidly registered with the 3D iUS image using image gradient information. The contour of the registered tumor model is visualized on the monitor of the navigation system. The rigid registration step was offline evaluated on 15 patients who overcame a brain tumor operation. The registered tumor models were compared with manual segmentations of the brain tumor in the 3D iUS data. Averaged DSI values of 82.3% and 68.4% and averaged contour mean distances of 1.7 mm and 3.3 mm were obtained for brain metastases and glioblastomas respectively. Future works will include the improvement of the functions in the pipeline, the integration of the pipeline into a centralized assistance system including further fonctionalities and connected with the navigation system, and the evaluation of the system during brain tumor operations.
Eurographics Workshop on Visual Computing for Biology and Medicine 2017: Frontmatter
(Eurographics Association, 2017) Bruckner, Stefan; Hennemuth, Anja; Kainz, Bernhard; Hotz, Ingrid; Merhof, Dorit; Rieder, Christian; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Maximizing AUC with Deep Learning for Classification of Imbalanced Mammogram Datasets
(The Eurographics Association, 2017) Sulam, Jeremias; Ben-Ari, Rami; Kisilev, Pavel; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Breast cancer is the second most common cause of death in women. Computer-aided diagnosis typically demand for carefully annotated data, precise tumor allocation and delineation of the boundaries, which is rarely available in the medical system. In this paper we present a new deep learning approach for classification of mammograms that requires only a global binary label. Traditional deep learning methods typically employ classification error losses, which are highly biased by class imbalance - a situation that naturally arises in medical classification problems.We hereby suggest a novel loss measure that directly maximizes the Area Under the ROC Curve (AUC), providing an unbiased loss. We validate the proposed model on two mammogram datasets: IMG, comprising of 796 patients, 80 positive (164 images) and 716 negative (1869 images), and the publicly available dataset INbreast. Our results are encouraging, as the proposed scheme achieves an AUC of 0.76 and 0.65 for IMG and INbreast, respectively.
Visual Navigation Support for Liver Applicator Placement using Interactive Map Displays
(The Eurographics Association, 2017) Hettig, Julian; Mistelbauer, Gabriel; Rieder, Christian; Lawonn, Kai; Hansen, Christian; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Navigated placement of an ablation applicator in liver surgery would benefit from an effective intraoperative visualization of delicate 3D anatomical structures. In this paper, we propose an approach that facilitates surgery with an interactive as well as an animated map display to support navigated applicator placement in the liver. By reducing the visual complexity of 3D anatomical structures, we provide only the most important information on and around a planned applicator path. By employing different illustrative visualization techniques, the applicator path and its surrounding critical structures, such as blood vessels, are clearly conveyed in an unobstructed way. To retain contextual information around the applicator path and its tip, we desaturate these structures with increasing distance. To alleviate time-consuming and tedious interaction during surgery, our visualization is controlled solely by the position and orientation of a tracked applicator. This enables a direct interaction with the map display without interruption of the intervention. Based on our requirement analysis, we conducted a pilot study with eleven participants and an interactive user study with six domain experts to assess the task completion time, error rate, visual parameters and the usefulness of the animation. The outcome of our pilot study shows that our map display facilitates significantly faster decision making (11.8 s vs. 40.9 s) and significantly fewer false assessments of structures at risk (7.4 % vs. 10.3 %) compared to a currently employed 3D visualization. Furthermore, the animation supports timely perception of the course and depth of upcoming blood vessels, and helps to detect possible areas at risk along the path in advance. Hence, the obtained results demonstrate that our proposed interactive map displays exhibit potential to improve the outcome of navigated liver interventions.
Mammogram Classification and Abnormality Detection from Nonlocal Labels using Deep Multiple Instance Neural Network
(The Eurographics Association, 2017) Choukroun, Yoni; Bakalo, Ran; Ben-Ari, Rami; Akselrod-Ballin, Ayelet; Barkan, Ella; Kisilev, Pavel; Stefan Bruckner and Anja Hennemuth and Bernhard Kainz and Ingrid Hotz and Dorit Merhof and Christian Rieder
Abstract Mammography is the common modality used for screening and early detection of breast cancer. The emergence of machine learning, particularly deep learning methods, aims to assist radiologists to reach higher sensitivity and specificity. Yet, typical supervised machine learning methods demand the radiological images to have findings annotated within the image. This is a tedious task, which is often out of reach due to the high cost and unavailability of expert radiologists. We describe a computeraided detection and diagnosis system for weakly supervised learning, where the mammogram (MG) images are tagged only on a global level, without local annotations. Our work addresses the problem of MG classification and detection of abnormal findings through a novel deep learning framework built on the multiple instance learning (MIL) paradigm. Our proposed method processes the MG image utilizing the full resolution, with a deep MIL convolutional neural network. This approach allows us to classify the whole MG according to a severity score and localize the source of abnormality in full resolution, while trained on a weakly labeled data set. The key hallmark of our approach is automatic discovery of the discriminating patches in the mammograms using MIL. We validate the proposed method on two mammogram data sets, a large multi-center MG cohort and the publicly available INbreast, in two different scenarios. We present promising results in classification and detection, comparable to a recent supervised method that was trained on fully annotated data set. As the volume and complexity of data in healthcare continues to increase, such an approach may have a profound impact on patient care in many applications.

Browse

Browsing VCBM 17: Eurographics Workshop on Visual Computing for Biology and Medicine by Issue Date

Results Per Page

Sort Options