Multi-core Accelerated Operation on Binary Volume ∗ Sheng-hui Liao, Yi-xiong Liang, Bei-ji Zou School of Information Science and Engineering Central South University Changsha, Hunan Province, China shliao@zju.edu.cn Keywords:Ray Casting, Segmented Binary Volume, Multi-layer-graph, Direct Volume Modification Abstract.This paper presents an efficient framework for direct volume rendering and manipulation of binary volume model. A data structure called multi-layer-graph, including a multi-layer volume and an assisted run-length encoding of boundary cells, is used to represent the model. Compared to common-used segmented binary volume, the multi-layer model enables the ray casting algorithm to produce much better image quality. Based on the structure feature of the data, techniques such as presence acceleration by boundary projection and rapid gradient estimation are employed to improve performance. For manipulation, basic removing and filling procedures are designed to support practical operations such as cutting, repairing and carving on the volume model, with interactive feedback in 3D scene. Additional improvement in performance can be achieved by multi-core acceleration. All these techniques have been integrated to a system for surgical simulation. 1. Introduction Computer-based surgical simulation has many applications in medical education, surgical training, surgical planning, which can reduce costs, provide experience with a greater variety of pathologies, and provide the ability to repeat or replay training procedures. Two important technologies required in surgical simulation are model rendering and manipulation, whose implementations are closely dependent on the underlying model representation. Surface-based models are frequently used, whose representation is compactly and the rendering is efficient by taking advantage of traditional hardware support. However, there are also a number of limitations. First, most medical data (e.g. CT, and MRI) is stored as a volumetric dataset, so surface model must first be extracted [1]. Second, as there is no interior structure, surface model can not easily support sophisticated manipulations such as free-form cutting, local repairing and carving. In other words, a spatially enumerative description that stores all interior information about the structure is necessary for manipulation, since we cannot tell which aspects of the structure －its interior or other structures in its vicinity －will contribute to the rendition after manipulation. These requirements imply the need for volumetric models. Instead, direct volume rendering is used a lot for displaying medical data sets, without first fitting geometric primitives. Since all voxels are contributing to the image, volume rendering displays much more interior phenomena by using transparency. On the other hand, it turns out to be much more computationally intensive than surface. In addition, object interior information of original volume is ∗ The research was supported by the National Natural Science Foundations (No.60903136, No.60970098) of China, the Special Postdoctoral Science Foundation of China (No.201003723), the Fundamental Research Funds for the Central Universities (No.201021200062). 2012 International Conference on Biological and Biomedical Sciences Advances in Biomedical Engineering, Vol.9 978-1-61275-027-9/10/$25.00 ©2012 IERI ICBBS 2012 49