Segmentation and quantification of oropharynx and larynx tumours from MRI data

Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2016
Thesis identifier
  • T14310
Person Identifier (Local)
  • 201269387
Qualification Level
Qualification Name
Department, School or Faculty
  • Radiation therapy (RT) is often offered as the primary treatment for the head and neck cancer. Quantitative analysis and volumetric measurements in RT require segmentation of a tumour (gross tumour volume (GTV)) and other anatomical structures (organs at risk). The current tumour segmentation technique, manual segmentation, using medical imaging is subject to high observer variability. Thus, this thesis describes new image processing methods for oropharynx and larynx tumours (segmentation and quantification) analysis from magnetic resonance imaging (MRI). An integrated approach has been developed, including data size and MRI artefacts reduction, throat region detection, extraction and segmentation oftumour (GTV) region with 3D reconstruction and quantification. Initially, a novel 2D automatic pharyngeal and laryngeal cancer segmentation framework (PLCSF) is proposed for oropharynx and larynx tumours segmentation from contrast enhanced T1-weighted axial MRI slices. In PLCSF, prior to segmentation, local entropy minimisation technique is employed to reduce intensity in homogeneity and new fuzzy rules based method is used for the throat region detection. Moreover a novel modified fuzzy c-means (FCM) clustering algorithm ispresented that is shown to robustly extract a tumour region compared to standard FCM clustering. Then a tumour segmentation boundary (outline) is obtained using region-based level set method after noise removal using non-linear filtering. The advantage of the proposed PLCSF lies in its ability to obtain a comparable tumour outlines even in presence of artefacts, heterogeneous tumour profile and fuzzy boundaries. Further an approach for three dimensional (3D) reconstruction and quantification of a tumour is presented. In this approach tumour outlines obtained from contiguous 2D slices are reconstructed in 3D using interpolation and rectangular mesh generation technique and volume is calculated using slice profile. Experimental results of PLCSF with volumetric measurements for oropharynx and larynx tumours on synthetic and real MRI data demonstrate that this tool may help reduce observer variations and can assist clinical oncologists with time-consuming,complex radiotherapy treatment planning. Finally, a novel automatic 3D throat region segmentation algorithm is presented. This algorithm efficiently combines Fourier interpolation and 3D level set segmentation technique to improve the accuracy of the segmentation results.
Resource Type