GPU accelerated stratified squamous epithelium biopsy image segmentation for OSCC detector and classifier

Abstract

Automatic cancer detection and classification is one of the open research problems. Oral squamous cell carcinoma (OSCC) is prevalent among oral cancer patients. Traditional procedures of detection and classification of biopsy specimen are tedious and clinico pathological acumen. Most of the computer aided biomedical image analysis algorithms suffer from slow speed due to central processing unit (CPU) based sequential implementation. Adapting parallel processing for such algorithms can improve operational speed of such algorithms. In this work, NVIDIA graphical processing unit (GPU) GeForce GTX 1050Ti is used to offload segmentation process and part of Laws texture feature calculations in stratified squamous epithelium biopsy image classifier (SSE-BIC) from CPU. SSE-BIC detects and classifies oral SSE images either normal or one of the three grades of malignancy. Image segmentation of SSE-BIC is implemented using GPU which includes 2-D convolution, principal component analysis (PCA) and k-means clustering. 2-D convolution employed for Laws texture features also implemented using GPU. In this way, CPU based serial executable classifier SSE-BIC is altered to accommodate parallel processing and compared with the CPU implementation. Results showed that parallel implementation is about 13.04X times faster than the serial CPU implementation of SSE-BIC.