Efficient brain lesion segmentation using multi‐modality tissue‐based feature selection and support vector machines

Abstract

Support vector machines (SVM) are machine learning techniques that have been used for segmentation and classification of medical images, including segmentation of white matter hyper-intensities (WMH). Current approaches using SVM for WMH segmentation extract features from the brain and classify these followed by complex post-processing steps to remove false positives. The method presented in this paper combines advanced pre-processing, tissue-based feature selection and SVM classification to obtain efficient and accurate WMH segmentation. Features from 125 patients, generated from up to four MR modalities [T1-w, T2-w, proton-density and fluid attenuated inversion recovery(FLAIR)], differing neighbourhood sizes and the use of multi-scale features were compared. We found that although using all four modalities gave the best overall classification (average Dice scores of 0.54 ± 0.12, 0.72 ± 0.06 and 0.82 ± 0.06 respectively for small, moderate and severe lesion loads); this was not significantly different (p = 0.50) from using just T1-w and FLAIR sequences (Dice scores of 0.52 ± 0.13, 0.71 ± 0.08 and 0.81 ± 0.07). Furthermore, there was a negligible difference between using 5 × 5 × 5 and 3 × 3 × 3 features (p = 0.93). Finally, we show that careful consideration of features and pre-processing techniques not only saves storage space and computation time but also leads to more efficient classification, which outperforms the one based on all features with post-processing.

Patch-Based Segmentation with Spatial Consistency: Application to MS Lesions in Brain MRI

“…Using a global threshold (TH) on a FLAIR image provides high sensitivity in the detection. Since such a threshold results in poor specificity, we can use the FLAIR input to provide an initial rough delineation of candidate lesion regions (see [ 26 ] for a detailed description). The lesions are characterized by demyelination; thus they are part of the WM tissue.…”

“…The final set of lesion candidate regions consists of all the voxels above the FLAIR TH intensity which are in the WM dilated region, as follows: where x is a voxel in the test image and TH is defined similar to [ 26 ] μ GM and σ GM are the mean and standard deviation of the GM tissue in the FLAIR test image, respectively. The parameter λ is an empirical parameter, selected experimentally as 0.5.…”

“…This stage significantly reduces the number of voxels that need to be further analyzed for lesions. It was found in [ 26 ] to decrease false positives (FPs), almost without any degradation of the true positives (TPs), that is, less than 5% loss. The parameter λ was chosen as the maximum value that maintains 95% of the lesions in the candidate lesion regions.…”

“…Typically the source of the training labels is manual segmentation. Typically studies use one of the many available supervised learning methods, for example, k -nearest neighbor ( k -NN) [ 19 ], support vector machine (SVM) [ 26 , 27 ], cellular neural networks (CNN) [ 28 ], or a Bayesian framework [ 29 ]. For example, one of the outstanding works [ 14 ] used a discriminative random decision forest to provide a voxel-wise probabilistic classification of the volume.…”

Special issue on computational methods for biomedical image processing and analysis

“…In this special issue on “ Computational Methods for Biomedical Image Processing and Analysis ”, two articles present procedures to segment brain images from MRI: E. Binaghi and V. Padoia present an automatic MRI 2D brain segmentation procedure based on graph‐searching techniques, and J.‐B. Fiot et al describe an efficient brain lesion segmentation algorithm using multi‐modality tissue‐based feature selection and support vector machines. Based on the finite element method, one biomechanical analysis is performed by W. Wolanski et al to study the craniosynostosis (a skull malformation) correction and another one by B. Gzik‐Zroska et al to assist the treatment of chest deformities.…”