CELL IMAGE CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORKS AND DIFFERENT IMAGE PREPROCESSING TECHNIQUES

Abstract

Medical image analysis field is highly dependent on good quality research that can result in time, cost improvements and aid in providing faster and better diagnosis for patients. Machine learning and especially convolution neural networks has proven to efficiently achieve the previously mentioned improvements in various medical field tasks. In this research we will focus on classification of cells based on their health level using a CNN model and several image preprocessing techniques with the goal of achieving high accuracy levels of predictions. The dataset used in this study has more than 20000 images for training and will be tested on two different datasets with each more than 8000 images. Several preprocessing techniques such as Wavelet denoising, Sobel filter, sharpening and edge enhancing filters will be tested and compared based on performance during the classification tasks with graphs and numerical results. The modified CNN model will be tested to find out the best parameters to use for training it and efficiently increasing the performance and precision.