Diagnosing brain tumours through functional imaging and machine learning

Zhao, Dadi Teddy ORCID: 0000-0001-7727-3638 (2023). Diagnosing brain tumours through functional imaging and machine learning. University of Birmingham. Ph.D.

Zhao2023PhD.pdf
Text - Accepted Version
Restricted to Repository staff only until 31 January 2032.
Available under License All rights reserved.
Download (9MB) | Request a copy

Abstract

Non-invasive in vivo functional imaging plays a key role in the clinical management of brain tumours. Magnetic resonance spectroscopy estimates the metabolite profiles of brain tissues. Diffusion weighted imaging shows the water diffusion map of the whole brain. Such imaging biomarkers facilitate the precise classification of brain tumours. Prior to classification, this the- sis optimised image and spectroscopy preprocessing and feature selection. Magnetic resonance spectroscopy was processed through adaptive wavelet noise suppression. Imaging features were selected based on the diagnostic ability across all tumour types. Additionally, imaging features of the minority class were oversampled through semi-synthetic wavelets oversampling. Combining these techniques, the cross validated classification accuracy for ependymomas, medulloblastomas and pilocytic astrocytomas was dramatically improved to be 100%. The findings from this thesis enhanced the role of magnetic resonance spectroscopy in clinical neuroscience.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):