Brain tumor cell recognition schemes using image processing with parallel ELM classifications on GPU

This paper investigates the possibility to enhance the recognition rate of brain tumor cell images acquired from the medical laboratory. A simplified image processing is first applied to the tumor cell as pre-processing, such as Otsu and unsharp masking methods; then, Histogram Orientation Gradient is our selection of feature extraction based on tumor shape characteristics which is then integrated into Extreme Learning Machine (ELM) as cell classification, called H-ELM. Its precision performance is confirmed against Support Vector Machine and a traditional ELM, i.e., 90% against 64% and 70%, respectively. To further improve H-ELM in aspects of computational complexity with high dimension and large image datasets, the feasibility to utilize the parallelism is investigated and implemented in classification stage using Compute Unified Device Architecture in Graphics Processing Unit resulting into 3 and 7 times speedup over its non-parallel scheme (CPU) and the traditional ELM, called Parallel H-ELM or PH-ELM.