Transfer-Deep Learning Application for Ultrasonic Computed Tomographic Image Classification

Abstract

Deep-learning techniques have led to a technological progress in several fields such as robotics, mechanics, and medicine specifically in the area of medical imaging. On the light of these recent developments in deep learning it will be recorded that medical imaging evolution needs a transfer deep learning application for the classification process. In this paper, our approach consists on a deep learning transfer models such us Inception V3, MobileNet, NasNet and Ameobanet on Ultrasonic Computed tomography images (USCT) to classify them automatically into three classes. In the beginning, USCT dataset augmentation has been done with pre-processing algorithms. Then, a Transfer Convolutional Neural Network Architecture has been applied with different models on our dataset. Finally, we have implemented our neural network application on GPU. As results we have overcoming previously works by a value of 100% for train accuracy and a value of 96% for test accuracy with a short time process.