Date of Award

Summer 8-9-2022

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Mathematics and Statistics

First Advisor

Yi Jiang

Second Advisor

Jun Kong

Third Advisor

Gengsheng Qin

Fourth Advisor

Yichuan Zhao

Abstract

Histopathological image classification has been at the forefront of medical research. We evaluated several deep and non-deep learning models for brain tumor histopathological image classification. The challenges were characterized by an insufficient amount of training data and identical glioma features. We employed transfer learning to tackle these challenges. We also employed some state-of-the-art non-deep learning classifiers on histogram of gradient features extracted from our images, as well as features extracted using CNN activations. Data augmentation was utilized in our study. We obtained an 82% accuracy with DenseNet-201 as our best for the deep learning models and an 83.8% accuracy with ANN for the non-deep learning classifiers. The average of the diagonals of the confusion matrices for each model was calculated as their accuracy. The performance metrics criteria in this study are our model’s precision in classifying each class and their average classification accuracy. Our result emphasizes the significance of deep learning as an invaluable tool for histopathological image studies.

DOI

https://doi.org/10.57709/30436059

File Upload Confirmation

1

Share

COinS