Application of deep learning techniques for biomedical data analysis

Khojasteh, P 2019, Application of deep learning techniques for biomedical data analysis, Doctor of Philosophy (PhD), Engineering, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Khojasteh.pdf Thesis application/pdf 4.08MB
Title Application of deep learning techniques for biomedical data analysis
Author(s) Khojasteh, P
Year 2019
Abstract Deep learning and machine learning methods have been used for addressing the problems in the biomedical applications, such as diabetic retinopathy assessment and Parkinson's disease diagnosis. The severity of diabetic retinopathy is estimated by the expert's examination of fundus images based on the amount and location of three diabetic retinopathy signs (i.e., exudates, hemorrhages, and microaneurysms). An automatic and accurate system for detection of these signs can significantly help clinicians to make the best possible prognosis can result in reducing the risk of vision loss. For Parkinson's disease diagnosis, analysis of a speech voice is considered as the earliest symptom with the advantage of being non-intrusive and suitable for online applications. While some reported outcomes of the developed techniques have shown the good results and ongoing progress for these two applications, designing new algorithms is a thriving research field to overcome the poor sensitivity and specificity of the outcomes as well as the limitations such as dataset size and heuristic selection of the network parameters.

This thesis has comprehensively studied and developed various deep learning frameworks for detection of diabetic retinopathy signs and diagnosis of Parkinson's disease. To improve the performance of the current systems, this work has had an investigation on different techniques: (i) color space investigation, (ii) examination of various deep learning methods, (iii) development of suitable pre/post-processing algorithms and (iv) appropriate selection of deep learning architectures and parameters.

For diabetic retinopathy assessment, this thesis has proposed the new color space as the input for the deep learning models that obtained better replicability compared with the conventional color spaces. This has also shown the pre-trained model can extract more relevant features compared to the models which were trained from scratch. This has also presented a deep learning framework combined with the suitable pre and post-processing algorithms that increased the performance of the system. By investigation different architectures and parameters, the suitable deep learning model has been presented to distinguish between Parkinson's disease and healthy speech signal.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Engineering
Subjects Biomedical Instrumentation
Keyword(s) Machine Learning
Deep Learning
Biomedical Data Analysis
Diabetic Retinopathy
Parkinson Disease
Version Filter Type
Access Statistics: 53 Abstract Views, 41 File Downloads  -  Detailed Statistics
Created: Wed, 11 Sep 2019, 14:10:41 EST by Adam Rivett
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us