The Classification of Retinal OCT Images Based on Convolutional Neural Network and Exploration

As deep learning algorithms developed rapidly in recent years, more and more classification and segmentation algorithms based on deep learning have been applied in the medical field due to their capacity to detect and extract disease features from medical images. This thesis reviews traditional and deep-learning image classification methods separately and makes comparisons between them. Using this analysis, a basic convolutional neural network (CNN) model was designed to be used as a deep-learning model to be used for processing image data. A publicly available retinal optical coherence tomography (OCT) dataset was employed to train and validate the model. The model was then refined and strengthened using theories of reducing overfitting, such as dropout, regularization, data augmentation and transfer learning. Eventually, the validation accuracy of the algorithm rose from under 20 percent to above 80 percent. In addition, the developed classifier model was then explored to segment tissue types in the images and produce a heat map of class activation to demonstrate additional clinical potential. Finally, this thesis discusses the challenges and practical strategies on CNN model for classification and segmentation.

Author

• Zhang, Ruixiang

Language

• en

Keyword

• Convolutional Neural Network
• Classification

Date created

• 2020

Type of Work

• Master's Thesis

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Biomedical Engineering

Degree

• M.S.

Advisor

• Zara, Jason

Committee Member(s)

• Loew, Murray
• Li, Zhenyu

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/c821gk41f