Please use this identifier to cite or link to this item:
Title: Survival Prediction of Glioblastoma Patients Using Pre-Operative Brain MRI Scans
Authors: Kaur, Gurinderjeet
Supervisor: Rana, Prashant Singh
Arora, Vinay
Keywords: Glioblastoma;Glioma Segmentation;Survival Prediction;Deep Learning;Machine Learning;Magnetic resonance imaging;Convolutional neural network;Radiomics
Issue Date: 15-May-2023
Abstract: Glioblastoma is identified as highly invasive and aggressive grade IV type glioma. Grade IV gliomas are malignant brain tumors that are cancerous by nature and cause life threatening consequences for the patients. The life span of glioblastoma patients is less than two years due to its abnormal growth and outspread to other parts of the patient’s body. Therefore, it is very important to predict the overall survival time of glioblastoma patients on prior basis for appropriate treatment planning. Early detection and precise segmentation of tumor region are two pre-requisites for accurate survival estimation of the patient. In this research work, retrospective study is done using Multimodal Brain Tumor Segmentation (BraTS) data of glioma patients made publicly available by the University of Pennsylvania. In first technique, the radiomic features were extracted using PyRadiomics Python library from pre-operative raw structural multi-parametric Magnetic Resonance Imaging (mpMRI) scans and, scans segmented using three-dimensional (3D) deep-supervised U-shaped Convolutional Neural Network (CNN) inspired encoder- decoder architecture after doing their necessary pre-processing. After removing irrelevant features, regression models based on Machine Learning (ML)were developed by considering selected radiomic features and clinical data to predict the Overall Survival (OS) time of Glioblastoma Multiforme (GBM) patients within a period of days only. In second technique, the features were extracted using a pre-trained two-dimensional (2D) Residual Neural Network (RNN) pre-operative raw structural mpMRI scans and, scans segmented using 3D deep-supervised UNet model architecture after doing their required preprocessing. The dimension-reduced principal components were integrated with clinical data and the handcrafted features of tumor subregions to compare the performance of automated ML-based regression techniques. In third technique, a lightweight 2D methodology was proposed to predict the survival time of GBM patients based on the pre-operative raw3D mpMRI scans and clinical data provided in the publicly available BraTS 2020 dataset. Firstly, a 2D Residual UNet for Segmentation (ResUNet-SEG) model was trained to perform semantic segmentation on brain tumour subregions. Then, the raw and segmented mpMRI volumes was used along with clinical data to train the 2D CNN for Survival Prediction (CNN-SP) model to predict the survival time in days.
Appears in Collections:Doctoral Theses@CSED

Files in This Item:
File Description SizeFormat 
Gurinderjeet - Thesis - Signed.pdfPhD Thesis27.99 MBAdobe PDFView/Open    Request a copy

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.