Performance Comparison of Parallel Computing Techniques for Processing CT Scan Images

In recent era, a shift is observed towards the use of GPUs instead of CPUs. This is due to the ever increasing processing power, excellent programming support and high performance to price ratio. Along with this, medical imaging is booming and it plays an important role in the diagnostics and treatment planning. However, the process of 3D image reconstruction is computationally very taxing and very data heavy. In this paper, two parallel computing platforms are investigated 1. Multi-core using OpenMP and, 2. Many-core using Compute Unified Device Architecture (CUDA) enabled Graphics Processing Unit (GPU). These approaches are evaluated based on the performance against sequential execution of image preprocessing function. For illustration purpose, a case study of conversion of 2D CT scan images to 3D reconstruction image is implemented on intel i3, 8th generation CPU with 4 cores using OpenMP. The same case study is implemented on CUDA enabled Nvidia Jetson TX2 GPU. We achieved speedup factor of 3 using OpenMP, whereas speedup factor of 107 using GPU for 1000 CT images.