Medical Image Segmentation using Deep Convolutional Neural Networks

Deep learning (DL) has been evolved in many forms in recent years, with applications not only limited to the Computer Vision tasks, expanded towards Autonomous Driving, Medical Imaging, Bio-Medical Imaging including Digital Pathology Image Analysis (DPIA), and in many other forms. Deep Convolutional Neural Network (DCNN) methods such as LeNet, AlexNet, GoogleNet, VGGNet, ResidulaNet, DenseNet, and CapsuleNet within the DL has been very successful in object classification and detection problems on a very large scale publicly available data set. Due to the great success of these DCNN methods, researchers have explored these methods to other imaging areas such as medical imaging problems, where there is a greater need for automated computer algorithms to make the diagnosis quick and cost-efficient, specifically for image classification, segmentation, detection, registration, and medical image data processing. Several state of art methods that provided superior performance in medical image segmentation such as Fully Connect Networks (FCN), SegNet, DeepLabs, U-Net, V-Net, and R2U-Net have outperformed hand-crafted machine learning algorithms. These models have been tested on several medical imaging and DPIA data sets but have not been explored on multi-organ segmentation, so the primary goal of this proposal is to explore more on these state of art models and test on several publicly available multi-organ segmentation data sets. The quantitative and qualitative performance will be evaluated against existing models using different performance metrics including, Accuracy, Sensitivity, Specificity, F1-score, Receiver Operating Characteristics (ROC) curve, dice coefficient (DC), and Mean Squared Error (MSE).https://ecommons.udayton.edu/stander_posters/2666/thumbnail.jp