IJSRSET173273 | Received : 06 March 2017 | Accepted : 20 March 2017 | March-April-2017 [(3)2: 196-209]
© 2017 IJSRSET | Volume 3 | Issue 2 | Print ISSN: 2395-1990 | Online ISSN : 2394-4099
Themed Section: Engineering and Technology
196
An Approach Patch-Based for the Segmentation of Pathologies
Application to Glioma Labelling
Syeda Meraj Bilfaqih, Sabahat Khatoon, Mariyam Ayesha Bivi
Department Of Computer Science, King Khalid University, Saudi Arabia
ABSTRACT
In this paper, we describe a novel and generic approach to address fully-automatic segmentation of brain tu-mors by
using patch-based voting techniques. In addition to avoiding the local search window assumption, the conventional
patch-based framework is enhanced through several simple procedures: an improvement of the training dataset in
terms of both label purity and intensity statistics, augmented features to implicitly guide the nearest-neighbor-search,
multi-scale patches, invariance to cube isometries, stratification of the votes with respect to cases and labels. A
probabilistic model auto-matically delineates regions of interest enclosing high-probability tumor volumes, which
allows the algorithm to achieve highly competitive running time despite minimal processing power and resources.
This method was evaluated on Multimodal Brain Tumor Image Segmentation challenge datasets. State-of-the-art
results are achieved, with a limited learning stage thus restricting the risk of overfit. Moreover, segmentation
smoothness does not involve any post-processing.
Keywords : Patch-Based, Glioma, Segmentation.
I. INTRODUCTION
A. Motivation
LIOBLASTOMA is the most severe case of brain
Gtumors. Clinical guidelines such as RECIST [1] or
RANO [2] are limited to 1D or 2D analysis (maximal
diameter and possibly second diameter) of the lesions.
However, from tumor growth monitoring to
radiotherapy planning, 3D anal-ysis is crucial in the
clinical pipeline [3], [4]. Glioblastoma segmentation
consists in a 3D delineation of the pathological
compartments [5] shown in Figure 1. Manual
segmentation is usually complex, subjective and time-
consuming. First, glioblastoma exhibit high tumor shape
variability. Second, the border between compartments
can appear fuzzy, which can lead to a debatable
segmentation: inter-rater variability of manual
segmentations is in the range 74-85% (Dice over-lap)
[5]. Third, the segmentation task requires the
simultaneous screening of 3D images acquired with
multiple Magnetic Res-onance (MR) sequences (Figure
1). This explains the ongoing interest for automatic
segmentation algorithms, notably within the Multimodal
Brain Tumor Image Segmentation (BraTS) benchmark
challenge [5].
Figure 1. MR channels (top row) ; whole brain
segmentation and mutually-inclusive pathological
regions (bottom row).