High Quality Volume Rendering For Medical Data

High quality volume rendering algorithm requires visualizing the regions of interest atan interactive frame rate. Ray casting direct volume rendering based on GPU meet the re-quirement for high quality volume rendering, but need transfer function to explore the innerfeatures of the dataset. Designing useful transfer function for direct volume rendering is atough job for users, and requires domain knowledge. A trial and error method is the mostused way for data classification with transfer function, which is time consuming and diffi-cult to get approving result. Maximum intensity difference accumulation is one of the fastdata feature explore methods, which combine the advantages of direct volume renderingand maximum intensity projection, but can’t visualize the local features within the dataset.In this paper, we present a local intensity difference accumulation volume rendering meth-od based on feature analysis, which can interactively visualizing the local features withinthe dataset.The algorithm we presented includes two stages: a) feature analysis for the volume da-taset; b) local intensity difference accumulation for features. The details are introduced asfollows.(1) Feature analysis. A Gaussian filter is employed to eliminate high frequency noiseduring the preprocess process. Moving least squares is introduced to reconstruct a continu-ous scalar curve of the ray profile. Then we search the ray profile for the local minimumpoints and introduce the user-defined gradient magnitude threshold to determine theboundary of different feature regions. Finally, we find the local features and classified thedataset.(2) Local feature enhanced volume rendering. Inspired by Maximum Intensity Differ- ence Accumulation (MIDA), a novel method Local Intensity Difference Accumulation(LIDA) is presented in this stage. We introduce depth reduction to improve the accumula-tion process and achieve better rendering result than MIDA. Surface shading technique isemployed to enhance the3D stereoscopic of the final rendering result. An adaptive loga-rithm Tone reduction technique is introduced to improve the local contrast of different fea-tures. Otherwise, an interaction method named feature threshold function is presented, thatuser can render different results interactively.The whole algorithm is implemented on the open source volume rendering plat-form-Voreen and GPU is used to accelerate the algorithm. Experiments show that our algo-rithm achieves interactive rendering efficiency and high quality results with apparent localfeatures. In this paper, we present several different experimental results on different da-tasets and proved that our algorithm is better in exploring local features of the volume da-taset than MIDA.