All ETDs from UAB

Advisory Committee Chair

Murat Tanik

Advisory Committee Members

Leon Jololian

Nakhmani Arie

Earl Wells

Karthikeyan Lingasubramanian

Document Type


Date of Award


Degree Name by School

Doctor of Philosophy (PhD) School of Engineering


Although Deep Leaning has achieved great success in many domains in recent years, research into its applicability and effectiveness in medical applications has been limited for various reasons. Some of these barriers are related to the historic impression that neural networks are black boxes, especially when applied to medical diagnosis and offer no in terpretation into prediction. In this dissertation, I demonstrated that deep learning can be an effective utility in medical diagnosis and biomarker detection. In particular, I have developed deep learning models that can identify biomarkers in brain hemorrhage, brain tumor, pneumonia, diabetic retinopathy and Parkinsons disease, with high sensitivity and specificity. I demonstrate models and methodologies in this thesis to address a number of current challenges in utilizing neural nets for medical applications, including (1) demonstra tion of methods and techniques for efficiently managing 3D and higher dimensional data, (2) demonstration of the utility of class-activation mapping for gleaning intuition into the bases of decision making within a given trained neural net, and (3) demonstration the utility of combining class activation mapping with augmented weak supervision techniques to improve the efficiency of training of neural nets and allow for more rapid development of clinically relevant algorithms. This thesis is organized through an exploration of appropriate method ologies and techniques applied to four independent, and disparate types of problems in the medical domain, involving two 2D datasets, one 3D dataset, and one higher dimensional (Diffusion-Weighted Imaging Magnetic Resonance Imaging) dataset. I close my discussion by proposing a general framework for developing and deploying deep learning systems into existing hospital infrastructure with considerations for privacy and security

Included in

Engineering Commons



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.