Deep Learning Applications in Medical Image and Shape Analysis

Abstract

Deep learning is one of the most rapidly growing fields in computer and data science in the past few years. It has been widely used for feature extraction and recognition in various applications. The training process as a black-box utilizes deep neural networks, whose parameters are adjusted by minimizing the difference between the predicted feedback and labeled data (so-called training dataset). The trained model is then applied to unknown inputs to predict the results that mimic human's decision-making. This technology has found tremendous success in many fields involving data analysis such as images, shapes, texts, audio and video signals and so on. In medical applications, images have been regularly used by physicians for diagnosis of diseases, making treatment plans, and tracking progress of patient treatment. One of the most challenging and common problems in image processing is segmentation of features of interest, so-called feature extraction. To this end, we aim to develop a deep learning framework in the current thesis to extract regions of interest in wound images. In addition, we investigate deep learning approaches for segmentation of 3D surface shapes as a potential tool for surface analysis in our future work. Experiments are presented and discussed for both 2D image and 3D shape analysis using deep learning networks.