A comparison study of deep CNN models for brain tumor MRI image classification

Throughout the life journey, brain health conditions may emerge and are identified by disruptions in normal brain growth and brain functioning. One of the disruption that may manifest as a neurological condition is brain tumor, which incidence is increasing in recent years. Since manual method of...

Full description

Saved in:
Bibliographic Details
Main Author: Muhammad Imran, Mohd Ridzuan
Format: Thesis
Language:English
English
English
Published: 2021
Subjects:
Online Access:http://eprints.uthm.edu.my/6999/1/24p%20MOHD%20RIDZUAN%20MUHAMMAD%20IMRAN.pdf
http://eprints.uthm.edu.my/6999/2/MOHD%20RIDZUAN%20MUHAMMAD%20IMRAN%20COPYRIGHT%20DECLARATION.pdf
http://eprints.uthm.edu.my/6999/3/MOHD%20RIDZUAN%20MUHAMMAD%20IMRAN%20WATERMARK.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Throughout the life journey, brain health conditions may emerge and are identified by disruptions in normal brain growth and brain functioning. One of the disruption that may manifest as a neurological condition is brain tumor, which incidence is increasing in recent years. Since manual method of classifying brain tumor images is tedious and can only be done at certain diagnostic centers, an alternative way of using deep learning technique to detect abnormal tissues in the brain is undertaken in this work. In this project, different architectures of Convolutional Neural Network (CNN) models namely AlexNet, Residual Network-18 (ResNet-18) and GoogLeNet were adopted and compared for their performance in classification of brain tumor Magnetic Resonance Imaging (MRI) images via two different approaches namely transfer learning method and modified method. The performance of the CNN models were evaluated and the most reliable model for the classification of brain tumor MRI images was determined. From the comparison done between the transfer learning method and modified method, transfer learning AlexNet showed the highest accuracy of 87.11% while modified ResNet-18 demonstrated the lowest accuracy of 76.09%. Based on observation of this project results and findings from other sources, for both approaches taken for brain tumor image classification, the tuning of hyperparameters used in training options (for transfer learning method only), the variation of image data augmentation to avoid overfitting and the depth of CNN topology influences the accuracy of the CNN models. The best CNN model for the classification of brain tumor MRI images with great accuracy was proven to be transfer learning AlexNet model.