Computing Degree Show 2018

Evaluating the Performance of Deeply Learned Filters in Brain Tumour Mutation Status Prediction based on Multi-parametric MR Images

There are many different factors that affect what steps are taken in the treatment of a brain tumour. This project looks into noninvasive tumour mutation status prediction which would give important information on the tumour behaviour and help the doctors with producing an informed and more accurate treatment plan without causing discomfort to patients.


In more detail, texture analysis of the MR Images and especially of Diffusion Kurtosis Images is used to predict the IDH mutation status of a tumour. In this project the performance of deeply learned filters as the texture detection method is evaluated and compared to the results given by manual MR8 filters. Two types of deeply learned filters are tested in the scope of this project: filters of pretrained and a self-trained convolutional neural network where the pretrained network is a VGG16 network trained by the creators of the model and the self-trained network was trained on MR Images.


Viivi Pursiainen

Supervisor: Dr Jianguo Zhang