Towards Secure and Intelligent Diagnosis: Deep Learning and Blockchain Technology for Computer-Aided Diagnosis Systems
Date
2021-03-25
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0002-5776-7929
Type
Thesis
Degree Level
Doctoral
Abstract
Cancer is the second leading cause of death across the world after cardiovascular disease. The survival rate of patients with cancerous tissue can significantly decrease due to late-stage diagnosis. Nowadays, advancements of whole slide imaging scanners have resulted in a dramatic increase of patient data in the domain of digital pathology. Large-scale histopathology images need to be analyzed promptly for early cancer detection which is critical for improving patient's survival rate and treatment planning. Advances of medical image processing and deep learning methods have facilitated the extraction and analysis of high-level features from histopathological data that could assist in life-critical diagnosis and reduce the considerable healthcare cost associated with cancer. In clinical trials, due to the complexity and large variance of collected image data, developing computer-aided diagnosis systems to support quantitative medical image analysis is an area of active research. The first goal of this research is to automate the classification and segmentation process of cancerous regions in histopathology images of different cancer tissues by developing models using deep learning-based architectures. In this research, a framework with different modules is proposed, including (1) data pre-processing, (2) data augmentation, (3) feature extraction, and (4) deep learning architectures. Four validation studies were designed to conduct this research. (1) differentiating benign and malignant lesions in breast cancer (2) differentiating between immature leukemic blasts and normal cells in leukemia cancer (3) differentiating benign and malignant regions in lung cancer, and (4) differentiating benign and malignant regions in colorectal cancer.
Training machine learning models, disease diagnosis, and treatment often requires collecting patients' medical data. Privacy and trusted authenticity concerns make data owners reluctant to share their personal and medical data. Motivated by the advantages of Blockchain technology in healthcare data sharing frameworks, the focus of the second part of this research is to integrate Blockchain technology in computer-aided diagnosis systems to address the problems of managing access control, authentication, provenance, and confidentiality of sensitive medical data. To do so, a hierarchical identity and attribute-based access control mechanism using smart contract and Ethereum Blockchain is proposed to securely process healthcare data without revealing sensitive information to an unauthorized party leveraging the trustworthiness of transactions in a collaborative healthcare environment. The proposed access control mechanism provides a solution to the challenges associated with centralized access control systems and ensures data transparency and traceability for secure data sharing, and data ownership.
Description
Keywords
Deep Learning, Medical Image Processing, Computer-Aided Diagnosis, Blockchain Technology
Citation
Degree
Doctor of Philosophy (Ph.D.)
Department
Computer Science
Program
Computer Science