Design Method Of Transfer Function For 3D Medical Data Volume Rendering

With the rapid development of modern medical imaging and computer technol-ogy, the visualization techniques for the three-dimensional medical data, such as direct volume rendering, are playing a more and more important role in the diagnosis and treatment. The visualization can help clinicians and researchers to see and understand the internal structures directly, especially the size of interested structures and the rela-tionship between different structures. Compared with the traditional two-dimensional slices, the medical data visualization provides users a more realistic display manner and quantitative analysis tools, helps users to diagnose the medical images and guide treat-ment.Direct volume rendering is often used to visualize medical data. The design of transfer function (TF) is one of the most important steps in the direct volume rendering. TFs are designed to map data properties (e.g. scalar value, gradient) to optical proper-ties (e.g. color, opacity). Color is used to generate a visual distinction between different data properties. Opacity determines the visual degree for each voxel of the volume data. TF design determines the quality of the visualization.This paper focuses on the research of the high efficient TF design and the inter-active methods. The aim is to avoid the complex interactive design and improve the efficiency of medical volume rendering. The classic TF designs for boundaries, such as f-|â–½f| histogram and LH histogram, can cluster and extract boundaries quickly. How-ever, under the conditions of noise, the clustering becomes blurred in the histograms, when the effect reaches a certain level, it is impossible for users to pick out the bound-aries with high quality. Due to the demand that users need to visualize the volume data from different viewpoints, it is necessary to adjust the TF to make the organizations or structures of interest highlighted. However, the adjustment of the volume rendering parameters is a trial-and-error and time-consuming process for the users with no prior knowledge. For these problems, this paper improves the above algorithms in the areas of TF design and adjustment of the volume rendering parameters.1. For the visualization of CT images, this paper proposes a TF design method based on the dynamic M-|â–½f| histogram. On one hand, we build a generalized bound-ary model contaminated by noise. On the other hand, based on the model we propose a novel multidimensional TF design method using boundary middle value (M), bound-ary height (â–³h) and gradient magnitude (|â–½f|) as data properties. Unlike traditional histograms presenting all boundary information once, the proposed method employs dynamic M -|â–½f| histogram and presents only one or a few boundaries at each step. A simple and iterative strategy of boundary extraction is also developed. We first sort different boundaries according to the values of their heights (â–³h) from high to low. Users control the â–³h value until a vertical bar occurs in M-|â–½f| histogram, and then pick out the boundary represented by the bar in M-|â–½f| histogram. Boundaries are extracted one by one until the â–³h value reaches 0. The misclassification among differ-ent boundaries can be reduced with the help of boundary ordering and the one-by-one extraction strategy. Besides, region elimination and region growing are further adopted to enhance the quality of rendering.2. This paper also proposes a TF design method though a what material you pick is what boundary you see approach. Unlike the traditional TF design based on histograms, we first propose three boundary visualization criteria and our boundary model. Com-pared with the classic two-material boundary model that is employed by LH histogram, our model only focuses on one material to achieve the human-centric interaction and in-troduce edge points for the precise localization of boundaries. Based on this boundary model we propose a what material you pick is what boundary you see boundary visual-ization that accepts direct manipulations on the original volumetric data, which enables the integration of semantics. Users can directly pick out the material of interest in the 2-D sequential images to transmit semantics. In addition, we utilize 3-D Canny edge detection to ensure the good detection and localization of boundaries. Furthermore, we establish a point-to-material distance measure to guarantee the accuracy and integrity of boundaries. Without the tedious region extraction in a newly defined multi-dimensional TF domain and concerning about the quality of the TF design, users can focus on the exploration of the volumetric data intuitively.3. It is tedious and time-consuming to fine-tune the opacity to achieve high quality of visualization. Even with high-quality TFs the rendering results may still contain am-biguous information so that users may be misled. Possible reasons are the loss of depth information when a 3-D volume compresses to a 2-D image and information superposi-tion in the form of occlusion while using ray casting or other rendering algorithms. On one hand, we propose the general notion of visibility ratio and occlusion vector. These two size-independent measures are physical quantities acting as the metric of occlusion. On the other hand, we propose a linear feedback mechanism that can be easily integrated into visualization systems based on data-centric TFs as a supplement. Users are liber-ated of fine-tuning parameters caused by occlusion and concentrate on the data-centric TF design with the aid of our proposed method. It is straightforward to integrate our method into other visualization systems based on the data-centric TFs as an important supplement.We apply our methods on different data sets to demonstrate the intuitive interaction and the effectiveness of the TF design and visualization.