CN104599311A - GPU (Graphics Processing Unit)-based hybrid visual system of three-dimensional medical image - Google Patents

Abstract

The invention realizes a GPU (Graphics Processing Unit)-based hybrid visual system of a three-dimensional medical image. The system adopts a three-dimensional texture mapping method to realize direct volume rendering, the programmable characteristic of the GPU is utilized to finish the transferring and the real-time revision of an interpolation post-classification algorithm and a transmission function in a volume rendering method, surface rendering is realized by an OpenGL technology, and the management of surface data is realized on the basis of a scene graph structure. Surface rendering and body rendering are both realized by the OpenGL, and the system realizes the hybrid displaying of the surface rendering and the body rendering by utilizing a fusion mechanism of the OpenGL. The system greatly improves the speed of the body rendering, effectively keeps the advantages of the surface rendering and the body rendering and enriches image information on the basis that rendering speed is guaranteed.

Description

Based on the 3 d medical images mixing visualization system of GPU

Technical field

A kind of field about the 3-D view technical elements in computer realm.

Background technology

From the 70's Mos of 20th century, many data field visualization method for drafting are suggested and are successfully applied to medical domain.Current many magic magiscans, such as three dimensional radiotherapy treatment planning system, virtual human body system, computer-asisted intra-operative navigation system, Medical Image Workstation etc., all contain the display section of 3-D view, but various piece re-constructs all separately.Some special VOLUME VISUALIZATION systems developed, because it does not have all functions of medical image three-dimensional display, lack again certain opening and extendibility simultaneously, are difficult to the instrument as secondary development.

The method for visualizing of 3 d data field mainly contains iso-surface patch and volume drawing two class].Iso-surface patch can the surface of drawing three-dimensional object effectively, but lacks the expression of internal information; Volume drawing is using voxel as elementary cell, directly generate final image by volume data, also direct volume drawing is claimed, it can represent the internal information of subject, but calculated amount is very large, although have employed many kinds of accelerating algorithm, the Real time dynamic display of medium-scale volume data (such as medical 3 D data) still can not be realized on common PC machine.In recent years, game and casino market huge promotion under, the technology constantly progress of PC graphic hardware aspect, performance improves constantly, many original only in the throe-dimensional temperature function that the throe-dimensional temperature hardware of high-grade graphics workstation just has, in being equipped with for common PC machine, high video card also possesses.Carry out the direct volume drawing of volume data by throe-dimensional temperature function, under the prerequisite reaching equivalent effect, greatly can improve drafting speed.In the utility system of medical domain, these two kinds of display packings are all necessary, two kinds of drawing result mixed display can be given full play to respective advantage, rich image information.Kreeger [2] etc. adopt hardware throe-dimensional temperature to achieve with Z2Buffer the drafting mixing scene, although can play up fast, owing to have employed P re2Shaded technology, thus cause resultant image quality not high.The programmable shader of a new generation's graphic process unit changes the drafting mode of GPU stationary pipes diatom in the past, makes programmer to intervene image generation process.

Summary of the invention

The mark of a new generation's graphic process unit there is provided the programmability of graphic hardware, is namely integrated with diabolo summit and carries out some unit's renderer (V ertex shader) of geometric transformation and illumination calculation and carry out painted sheet unit's renderer (F ragm en t shader or P ixel shader) to pixel.Summit pipeline in the completely not alternative conventional processors of programmable shader and the fixing processing unit of pixel pipeline.

In order to utilize the programmability of Modern Graphic processor, Graphic programmers needs to be programmed in the program that graphic process unit is run, and this program is called coloring process (Shader).Graphic process unit provides programmability by general 3D A P I (D irectX O penGL) to Graphic programmers.In order to be consistent with iso-surface patch part, adopt OpenGL Interface realization coloring process herein.

Workload most important and maximum in volume rendering is sampled and interpolation calculation exactly.Due to the discrete data obtained after volume data inherently a kind of sampling, this process is also referred to as resampling.In order to draw out a high-quality volumetric image, the corresponding numerical value of these resample points needs carefully to calculate.Using volume data as three-D grain, the texture interpolation function of graphic hardware can be utilized, obtain the numerical value of resample points fast.

The classification of 3 d data field refers to and transition function is acted on volume data, and donicity composes color value and opaque value according to each point.Classification is a very important step in whole visualized algorithm, only has and carries out correct classification to three-dimensional data value, could draw rational image through subsequent treatment.[7] are classified after classification and interpolation before sorting technique mainly contains interpolation.Based in the direct volume drawing of three-D grain, if sorting technique before employing interpolation, as three-D grain is the color value and opacity value of having classified to volume data, not only needs storage space (R, the G of 4 times, B, A tetra-passages, i.e. volume data size 3 4), and during sampling interpolation to as if color value and opacity value, instead of original volume data value, final picture quality [7] can be reduced.After interpolation, classification needs using original volume data as three-D grain, and the object of sampling is like this exactly initial body data, participates in Images uniting again after sampled point classification.

After interpolation, classification needs after extraction texture again to its action transport function, obtain RGBA value corresponding with it, this cannot realize in original fixing graph rendering pipeline, because programmer cannot insert a step and send into frame buffer participation fusion again after texture sampling, Gefo rce 6800 graphics card used herein has programmable functions, utilizes programmable functions can realize sorting technique after interpolation easily.

The sheet unit processor of graphics card has texture blending instruction, and after interpolation, classification can rely on to extract with texture and realize, and texture relies on and extracts schematic diagram as shown in Fig. 3.Value by the 1st texture extracts texture as the coordinate of next texture, therefore can using three-dimensional data texture as the 1st texture, transition function represents in the mode of the 2nd one dimension texture (one dimension transition function) or 2 d texture (two-dimentional transition function), and transition function represents in 2 d texture mode.The value of the 1st texture extracted like this is the value after texture interpolation, goes extraction the 2nd texture to be equivalent to action transport function and classifies, thus classify after achieving interpolation with this value.Write P ixelshader with GL SL, its code is vec4 index=tex tu re3D (m y3DTex tu re, gl-

TexCoo rd[0 ]) 。

gl- F ragCo lo r= tex tu re2D (Dependen tTex tu2 re, index. x, index. y) 。

Transition function is all the major issue of direct volume drawing all the time.It mainly have mapped texture scalar value sto the colored forms of RGBA, it is the key determining some important feature and details display.The real-time, interactive amendment that texture can realize transition function is relied on by amendment.

One, surface extraction;

Surface rendering is exactly by extracting section interested in data fields out and express in the mode of dough sheet, and conventional is contour surface.The method extracting contour surface is divided into two classes: (1) slice-level method for reconstructing, namely first in a series of section, generates isoline, then connects isoline and obtains contour surface; (2) voxel level method for reconstructing, namely directly extracts contour surface from volume data.1st class methods are a lot of owing to generating isoline method in section, so dirigibility is good, but isoline is corresponding and splicing and all more difficult to the process of bifurcated situation.The most representative M arch ing Cubes algorithm being document [8] and proposing in 2nd class methods.

Two, organization and administration;

Tissue surface can adopt multiple method to extract, and can extract some tissues from an individual data items, so correctly show according to its relative coordinate seem extremely important to the management of these tissues.Scene graph provides a kind of well data management mode.Scene graph refers to a kind of data structure, and scene graph is directed acyclic graph, is formed by connecting by each node.The node of scene graph is divided into group node and leaf node, and group node is the skeleton of whole scene graph, and leaf node then contains the specifying informations such as body object, attribute.Whole scene graph only has a root node, represents whole volume data, and its child node of each setup action extracted.If some setup action entirety, under these tissues can being hung over same group node, using this group node as under a child node puts into root node.Based on these features of scene graph, the visualization system application scenarios figure designed herein carries out organization and administration.

Three, the combination of iso-surface patch and volume drawing;

Iso-surface patch can reflect the three-dimensional structure of complex object compactly, but Surface rendering techniques draws is the surface of extracting from volume data, the information that can not comprise in comprehensive representation three-dimensional data; Volume drawing directly calculates last image from volume data, but draw based on object plotting method the normal information that contour surface needs surface, the shade of gray of volume elements is stored in the RGB component in three-D grain data, gray-scale value is stored in A lpha component, then utilize the A lpha test function of hardware supported to show contour surface, this method is very effective, but needs to store than simple the video memory space consuming that gray-scale value has more 3 times.Iso-surface patch method is adopted to draw contour surface herein, adopt other information of object plotting method expression body data, the programmable features of GPU is utilized two kinds of image blend to be shown, do not need like this to store Grad in three-D grain data, only need to store gray-scale value, save video memory space, and can simultaneously display surface and internal information.The method for drafting that this iso-surface patch and volume drawing combine, can retain the advantage of iso-surface patch and volume drawing effectively.

Ultimate principle based on throe-dimensional temperature volume drawing is the A lpha hybrid technology of application video card, to the fusion acting on behalf of geometric surface sampled data and carry out from back to front (or from front to back), thus the approximate integral function realized in direct volume drawing algorithm.Iso-surface patch method is then adopt tri patch to approach contour surface, utilizes the pipeline of fixedly playing up of video card to carry out surface rendering.Because the OpenGL state parameter of these two kinds of method for drafting when drawing is different, directly carry out blend rendering may cause only having a kind of drawing result correctly to show or do not have Image Rendering to screen, the programmable features of the Render2to2tex tu re function and GPU of applying video card herein addresses this problem.Render2to2tex tu re function utilizes Render to target to expand and realizes, and exactly drawing result is outputted to texture cell, thus as the input texture of sheet unit renderer, again can participate in computing.

Claims (5)

1. it is characterized in that the direct volume drawing based on GPU based on the 3 d medical images mixing visualization system of GPU, based on the volume drawing of throe-dimensional temperature, classify after interpolation.

2. be exactly by extracting section interested in data fields out and express in the mode of dough sheet according to claim 1 surface rendering, conventional is contour surface.

3. the method extracting contour surface according to claim 2 is divided into two classes: (1) slice-level method for reconstructing, namely first in a series of section, generates isoline, then connects isoline and obtains contour surface; (2) voxel level method for reconstructing, namely directly extracts contour surface from volume data.

4. a lot of owing to generating isoline method in section according to claim 3 item, so dirigibility is good, but isoline is corresponding and splicing and all more difficult to the process of bifurcated situation.

5. be most M arch ing Cubes algorithm according to claim 3 item.