A novel magnetic resonance diffusion tensor imaging segmentation and visualisation technique for brain tumour diagnosis and surveillance

BACKGROUND Over 120 brain tumour subtypes have been identified, varying in presentation, treatment and life expectancy. Imaging has a role in diagnosis, treatment planning and evaluating therapeutic response. Advanced techniques e.g. diffusion tensor imaging (OTI) have been used to quantify voxel ma...

Full description

Bibliographic Details
Main Author: Jones, Timothy L.
Published: University of London 2012
Subjects:
Online Access:http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.589799
Description
Summary:BACKGROUND Over 120 brain tumour subtypes have been identified, varying in presentation, treatment and life expectancy. Imaging has a role in diagnosis, treatment planning and evaluating therapeutic response. Advanced techniques e.g. diffusion tensor imaging (OTI) have been used to quantify voxel magnitudes of isotropic (P) and anisotropic (q) diffusion. It has been proposed these reflect tissue structure and may characterise tumour type as well as delineate margins. METHODS Using two 1.5T MRI scanners, OTI scans were acquired from 94 patients prior to surgery and histological diagnosis (38 glioblastoma, 19 low-grade glioma (LGG), 26 metastasis and 11 meningioma). Furthermore, OTI scans were acquired from 6 LGG patients at 3 time points. Manual tumour and oedema regions of interest (MROI) were drawn on coregistered enhanced Tj-weiqhted and FLAIR MRI. Values of p and q within MROI were entered into a discriminant analysis. A novel k-means segmentation (k=16) of p:q space was performed simultaneously across all data sets and resultant clusters labelled according to constituent p, q and T 2-weighted characteristics, generating colour-diffusion-maps. Using a flood-filling technique, segmented ROls (SROI) were determined. Constituent segment profiles (spectra) were evaluated using the same discriminant analysis. SROI were created from longitudinal LGG scans and spectral changes evaluated over time. RESULTS Despite marked differences in p and q between MROls, the discriminant model revealed poor diagnostic accuracy. The segmentation is computationally efficient, stable and presents potential tumour-specific patterns with a delineated' boundary between tumour and brain. The SROI analysis provided an overall diagnostic sensitivity of 86%. LGG spectra identified early features of malignant transformation not evident on conventional MRI. CONCLUSIONS We present a novel visualisation of tumours from OTI and a diagnostic tool with good diagnostic accuracy and ability to identify changes in LGG. It delineates a margin which may represent a clinically relevant boundary for treatment planning and volumetric analysis.