PC Platform Based Medical Imaging System

3D data fields are a hot research area in scientific visualization, which are also met generally in practical applications. Among all the visualization methods, direct rendering of 3D data fields is a very important approach. Compared with the isosurface extraction approach, direct volume rendering processes all samples to make images without producing the intermediate geometric elements. Therefore, many contents can be displayed together in a single image, which benefits scientists to comprehend data.Ray casting algorithm can generate high-quality images, so it becomes the ideal solution of medical image visualization . However, the calculation of its huge volume data limits the display speed. This paper presents a software based acceleration algorithm: shear - warp algorithm. The algorithm decompose the 3D projection transformation to two steps: so that changing the process of re-sampling in 3D space to 2D space, greatly reducing the amount of computation. This makes the real time display of three-dimensional data sets in volume rendering graphics workstation possible, not significantly reducing image quality.With analysis of ray casting algorithm, it is not difficult to find the inherent parallelism among rays. After the comparison of the SIMD Extensions of Intel and AMD, this paper presents a SIMD Extensions based accelerated volume rendering method. The principal idea is to load the four parameters into the 128 bits register, which can be calculated in parallel. Trilinear interpolation in volume rendering is a time-consuming step, This paper also proposed a SIMD optimization for it, reducing the addition and multiplication times. In color cumulation process, to reduce the frequency of branch instruction, we us a SIMD saturation commands to limit the value above the maximum.Based on above algorithm, we develop a medical image visualization system which integrats hardware-based graphics and SIMD Extensions based solution. The system can read variety format of medical data, and produce different quality images depend on the user . In addition, the system provides real-time transfer function and perspective adjusting interface, so that users can observe different layers and part of volume data.Finally, this paper analyzes the performance of the system, presents that the SIMD optimization's bottleneck is LUT table search, while GPU is not fit for large dataset.