Maria Ferre

A framework for fusion methods and rendering techniques of multimodal volume data

Many different direct volume rendering methods have been developed to visualize 3D scalar fields on uniform rectilinear grids. However, little work has been done on rendering simultaneously various properties of the same 3D region measured with different registration devices or at different instants of time. The demand for this type of visualization is rapidly increasing in scientific applications such as medicine in which the visual integration of multiple modalities allows a better comprehension of the anatomy and a perception of its relationships with activity. This paper presents different strategies of direct multimodal volume rendering (DMVR). It is restricted to voxel models with a known 3D rigid alignment transformation. The paper evaluates at which steps of the rendering pipeline the data fusion must be realized in order to accomplish the desired visual integration and to provide fast re‐renders when some fusion parameters are modified. In addition, it analyses how existing monomodal visualization algorithms can be extended to multiple datasets and it compares their efficiency and their computational cost. Copyright

PREVIRNEC: A cognitive telerehabilitation system based on Virtual Environments

In this paper, we describe PREVIRNEC, a distributed system for cognitive telerehabilitation based on virtual environments. Our system allows personalized treatments by means of 2D and 3D exercises that can be built according to single patients characteristics. Patients realize their exercises remotely. Depending on their obtained results, the system readjusts automatically the levels of difficulty of the tasks, and switches from one task to the other. The paper focuses on technological issues of the system design. We analyze its structure, components, and we discuss the decisions adopted in the interaction mode design and well as in the tasks layout.

Decision trees for accelerating unimodal, hybrid and multimodal rendering models

This paper deals with the rendering of segmented unimodal, hybrid and aligned multimodal voxel models. We propose a data structure that classifies the segmented voxels into categories, so that whenever the model has to be traversed, only the selected categories are visited and the empty and non-selected voxels are skipped. This strategy is based on: (i) a decision tree, called the rendering decision tree (RDT), which represents the hierarchy of the classification process and (ii) an intermediate run-length encoding (RLE) of the classified voxel model. The traversal of the voxel model given a user query consists of two steps: first, the RDT is traversed and the set of selected categories computed; next, the RLE is visited, but the non-selected runs are skipped and only the voxels of the original model that are codified are accessed in selected runs of the RLE. This strategy has been used to render a voxel model by back-to-front traversal and splatting as well as to construct 3D textures for hardware-driven 3D texture mapping. The results show that the voxel model traversal is significantly accelerated.

Ray-casting time-varying volume data sets with frame-to-frame coherence

The goal of this paper is the proposal and evaluation of a ray-casting strategy that takes advantage of the spatial and temporal coherence in image-space as well as in object-space in order to speed up rendering. It is based on a double structure: in image-space, a temporal buffer that stores for each pixel the next instant of time in which the pixel must be recomputed, and in object-space a Temporal Run-Length Encoding of the voxel values through time. The algorithm skips empty and unchanged pixels through three different space-leaping strategies. It can compute the images sequentially in time or generate them simultaneously in batch. In addition, it can handle simultaneously several data modalities. Finally, an on-purpose out-of-core strategy is used to handle large datasets. The tests performed on two medical datasets and various phantom datasets show that the proposed strategy significantly speeds-up rendering.

Breast Masses Identification through Pixel-Based Texture Classification

Mammographic image analysis plays an important role in computer-aided breast cancer diagnosis. To improve the existing knowledge, this paper proposes a new efficient pixel-based methodology for tumor vs non-tumor classification. The proposed method firstly computes a Gabor feature pool from the mammogram. This feature set is calculated through multi-sized evaluation windows applied to the probabilistic distribution moments, in order to improve the accuracy of the whole system. To deal with a high dimensional data space and a large amount of features, we apply both a linear and non-linear pixel classification stage by using Support Vector Machines (SVMs). The randomness is encoded when training each SVM using randomly sample sets and, in consequence, randomly selected features from the whole feature bank obtained in the first stage. The proposed method has been validated using real mammographic images from well-known databases and its effectiveness is demonstrated in the experimental section.

A Fast hierarchical traversal strategy for multimodal visualization

In the last years there is a growing demand of multimodal medical rendering systems able to visualize simultaneously data coming from different sources. This paper addresses the Direct Volume Rendering (DVR) of aligned multimodal data in medical applications. Specifically, it proposes a hierarchical representation of the multimodal data set based on the construction of a Fusion Decision Tree (FDT) that, together with a run-length encoding of the non-empty data, provides means of efficiently accessing to the data. Three different implementations of these structures are proposed. The simulations results show that the traversal of the data is fast and that the method is suitable when interactive modifications of the fusion parameters are required.