Accelerating Convolutional Neural Network Training for Colon Histopathology Images by Customizing Deep Learning Framework

Abstract

Cancer diagnose based on the histopathology images is still have some challenges. Convolutional Neural Network (CNN) is one of deep learning architecture that has widely used in medical image processing especially for cancer detection. The high resolution of images and complexity of CNN architecture causes cost-intensive in the training process. One way of reducing the training processes time is by introducing parallel processing. Graphics Processing Unit (GPU) is a graphics card which has many processors and has been widely used to speed-up the process. However, the problem in GPU is the limitation of memory size. Therefore, this study proposes alternative ways to utilize the GPU memory in the training of CNN architecture. Theano is one of middle-level framework for deep application. GPU memory is a critical task in training activity and will affect to the number of batch-size. Customizing memory allocation in Theano can be conducted by utilizing library called ‘cnmem’. For training CNN architecture, we use NVIDIA GTX-980 that accelerated by customizing CUDA memory allocation from ‘cnmem’ library located in ‘theanorc’ file. In the experiment, the parameter of cnmem are chosen between 0 (not apply cnmem) or 1 (apply cnmem). We use image variation from 32x32, 64x64, 128x128, 180x180 and 200x200 pixels. In the training, a number of batch-size is selected experimentally from 10, 20, 50, 100 and 150 images. Our experiments show that enabling cnmem with the value of 1 will increase the speed-up. The 200x200 images show the most significant efficiency of GPU performance when training CNN. Speed-up is measured by comparing training time of GTX-980 with CPU core i7 machine from 16, 8, 4, 2 cores and the single-core. The highest speed-up GTX-980 obtained with enabling cnmem are 4.49, 5.00, 7.58, 11,97 and 16.19 compare to 16, 8, 4, 2 and 1 core processor respectively