Automated Glioblastoma Segmentation Based on a Multiparametric Structured Unsupervised Classification

Abstract

Automatic brain tumour segmentation has become a key component for the future of brain tumour treatment. Currently, most of brain tumour segmentation approaches arise from the supervised learning standpoint, which requires a labelled training dataset from which to infer the models of the classes. The performance of these models is directly determined by the size and quality of the training corpus, whose retrieval becomes a tedious and time-consuming task. On the other hand, unsupervised approaches avoid these limitations but often do not reach comparable results than the supervised methods. In this sense, we propose an automated unsupervised method for brain tumour segmentation based on anatomical Magnetic Resonance (MR) images. Four unsupervised classification algorithms, grouped by their structured or non-structured condition, were evaluated within our pipeline. Considering the non-structured algorithms, we evaluated K-means, Fuzzy K-means and Gaussian Mixture Model (GMM), whereas as structured classification algorithms we evaluated Gaussian Hidden Markov Random Field (GHMRF). An automated postprocess based on a statistical approach supported by tissue probability maps is proposed to automatically identify the tumour classes after the segmentations. We evaluated our brain tumour segmentation method with the public BRAin Tumor Segmentation (BRATS) 2013 Test and Leaderboard datasets. Our approach based on the GMM model improves the results obtained by most of the supervised methods evaluated with the Leaderboard set and reaches the second position in the ranking. Our variant based on the GHMRF achieves the first position in the Test ranking of the unsupervised approaches and the seventh position in the general Test ranking, which confirms the method as a viable alternative for brain tumour segmentation.

Comparison of unsupervised classification methods for brain tumor segmentation using multi-parametric MRI

“…So far only few studies have considered MP-MRI for brain tumor segmentation, although it has been commonly suggested in literature (Juan-Albarracín et al, 2015, Bauer et al, 2013, Dhermain, 2014). We have shown in a previous study that significantly higher segmentation accuracy could be achieved by applying hNMF to MP-MRI data compared to using only cMRI data (Sauwen et al, 2015).…”

“…Juan-Albarracin et al have recently combined some dedicated cMRI pre-processing steps with several unsupervised methods, including FCM, GMM and hidden Markov random fields. Their results ranked among the best supervised algorithms in the BRATS challenge (Juan-Albarracín et al, 2015). A direct comparison between our results and those from the BRATS challenge is not in place, as we are combining cMRI data with PWI, DWI and MRSI in a multi-parametric approach, whereas BRATS only considers cMRI data.…”

“…Therefore, further research will focus on automation by incorporating prior knowledge, e.g. tissue probability maps based on an image atlas (Juan-Albarracín et al, 2015), feature-specific knowledge (Kazerooni et al, 2015) or user input (Menze et al, 2015). …”

“…Mostly, each tissue type is modelled by a multi-variate Gaussian distribution, resulting in a Gaussian mixture model (GMM) (Menze et al, 2010, Zhang et al, 2001). Recently, GMM has been shown to be competitive with state-of-the-art supervised classification algorithms for GBM segmentation based on cMRI data (Juan-Albarracín et al, 2015). Over the last years graph-cut based methods have become popular as well.…”

“…Nearly all of the proposed segmentation algorithms have so far mainly been applied to cMRI data (Phillips et al, 1995, Fletcher-Heath et al, 2001, Menze et al, 2010, Zhang et al, 2001, Juan-Albarracín et al, 2015). However, along with the recent emergence of advanced MRI modalities such as diffusion-weighted MRI (DWI), perfusion-weighed MRI (PWI) and magnetic resonance spectroscopic imaging (MRSI), numerous studies have suggested that tumor characterization might benefit from the additional structural, biological and biochemical information provided by these advanced MRI modalities (Bauer et al, 2013, Cha, 2006, Padhani and Miles, 2010).…”