Date of Award

6-1-2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Systems Engineering

First Advisor

Mariofanna Milanova

Abstract

Internet of Things (IoT) technology is becoming increasingly common in the healthcare industry. The significant benefits of IoT are decreased operational costs, improved outcomes of the treatment through virtual infrastructures and accessibility of real-time information for making informed decisions, improved disease management, reduced errors, and enhanced patient experience. This enables hospitals to address the demands of the ever-growing population in this world. On the other hand, medical signal classification systems are systems that recognize normal from abnormal medical signals. These systems depend on two modes, training mode in which the system learns about the dominant features of the signals, and testing mode in which the system recognizes an unknown new signal. To accomplish the main aim, various objectives are categorized based on latency, accuracy, and classification speed. These objectives are as follows. First, create a new invariants method to extract the features of the action based on Hu moments technique to represent the dominant features of Alzheimer disease brain images, normalize these moments, and use two classifiers (K-Nearest Neighbors algorithm (KNN) and Linear Support Vector Machines (SVM)) to achieve highest possible accuracy. Second, a deep neural network using the (Caffe) framework with three different models (LeNet, AlexNet, and GoogLeNet) is applied for training dataset samples of electroencephalography (EEG) signals to accomplish optimal accuracy within no time. Next, introducing a new methodology for Alzheimer's disease classification based on TensorFlow Convolutional Neural Network (TF-CNN). Two main contributions have been done: data augmentation and using several optimizers (Adagrad, Proximal Adagrad, Adam, and RMSProp). The training dataset images are augmented by normalizing, rotating, and cropping them. The data augmentation helps to decrease overfitting and increase model performance. In our fourth work, an object localization for skin lesion detection has been proposed using Single Shot Detector- Mobilenet (SSD- Mobilenet) model on ISIC 2018 datasets (International Skin Imaging Collaboration) for training and testing modes. The detection process has been achieved using two different methods: a real-time mobile application of Android camera (Galaxy S6), and Jupyter Notebook of TensorFlow Object Detection Application Program Interface (API). Eventually, our research presents an approach for connecting and implementing three modules in a Raspberry pi3 device with the Custom Vision cloud services for classifying skin disease lesions. The modules are run on a Raspberry pi3 using Azure at the edge where the custom vision services exist. Two different datasets are used in this work, ISIC 2019 and DermIS.net, and performance assessment of IoT messaging protocols (HTTP (Hypertext Transferring Protocol), MQTT (Message Queue Telemetry Transport)) has been recorded.

Download

Share

COinS