Brain MRI Tumor Analysis and Classification using Deep Learning
Magnetic Resonance Imaging (MRI) is a medical imaging technique frequently used to produce a pictorial view of internal body parts, structure and functionality. Radiologists usually perform manual analysis on a large number of MR images for brain tumor detection. It is very hard to correctly segment...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/36104/6/Ghazanfar.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Magnetic Resonance Imaging (MRI) is a medical imaging technique frequently used to produce a pictorial view of internal body parts, structure and functionality. Radiologists usually perform manual analysis on a large number of MR images for brain tumor detection. It is very hard to correctly segment the tumor tissues present in the MR images due to the similarity, noise, complex texture, poor sampling and image distortions. With the advent of more powerful computing devices, system automation plays a pivotal role. In the medical industry, automated image classification and segmentation is an important task to assist medical experts in the correct diagnosis of a certain disease. Most of the recent research studies focus on binary classification of brain MR image into tumorous and non-tumorous images in addition to tumor segmentation. The extracted tumor is further classified into four Glioma tumor classes which are Necrosis, Edema, Enhancing and Non-Enhancing tumors. This is referred as multiclass classification, and it is an area that is yet to be explored. The current classification techniques proposed in the literature suffer from limited dataset size and overfitting problems. In this thesis, an automated multistage technique is proposed for the MR image classification into tumorous and non-tumorous images followed by brain tumor segmentation and classification in order to differentiate the tumors into these four classes using enhanced deep learning models. The experiments were performed using BraTS, AANLIB and PIMS-MRI datasets. In the first stage, the contribution of this study is a binary brain MR image classification, where two enhanced techniques are proposed to classify the images into tumorous and non-tumorous. The enhanced techniques are based on binary CNN features and binary Deep CNN classifier that resulted an average accuracy of 97.61% and 98.04%, respectively. In the second stage, the tumor region was segmented from the tumorous images using the proposed neighboring Fuzzy C-means (FCM) based technique which resulted an average dice similarity coefficient (DSC) score of 90.87%. In the third stage, the segmented tumor was classified into four Glioma tumor classes using an enhanced multiclass CNN features and Deep CNN model as classifier. The experimental results using CNN features and Deep CNN model as classifier for multiclass Glioma tumor classification into Necrosis, Edema, Non-enhancing tumor, and enhancing tumor indicated an average accuracy of 96.19% and 96.30% respectively, which outperformed results reported in previous literature. |
|---|