Tumor regions from normal tissues such as gray

Tumor is an uncontrolled growth ofunnatural cells in any part of the body 38. Tumor affects the brain tissuesand makes unpredictable abnormal cell development is called as brain tumor. Earlydetection and diagnosis of brain tumor is very much important to improving thetreatment planning and increases survival rate of the patients 48. Generallybrain tumor classified into primary brain tumors and malignant brain tumors. Primarybrain tumor start and spread into brain itself whereas malignant initial originmaybe anywhere in the body and start to spread inside the brain. The most common primary brain tumor iscalled Gliomas, which originate from Glial cells and infiltrate the surroundingtissues.

Majority of brain tumor segmentation methods are focus on Gliomasbecause of Gliomas are mostly targets the adults. Segmentation of Gliomas isvery difficult task due to variety of shapes, sizes and any position in thebrain lecture note 75 page. MRI is non-invasive techniques provide a goodcontrast on soft tissues using radio frequency and powerful magnetic field. MRIgenerates stack of 2D slices in multisequence format based on alteringexcitation and repetition time.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

The multisequence used for the Gliomasdiagnosis are: T1-weighted (T1), T1-weighted with gadolinium contrast enhancement(T1c), T2-weighted (T2) and Fluid Attenuated Inversion Recovery (FLAIR) 46.Sample multisequence images are shown in Fig.1.Brain tumor segmentation consists ofsegmenting active tumor, edema, and necrosis regions from normal tissues suchas gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) and shownin Fig.1. Manual segmentation of these regions from huge data is a timeconsuming painstaking task to the clinicians and prone to substantialinter-rater error.

To remove all these difficulties, automatic methods andalgorithms for classification, segmentation and visualization along with themodern systems provided a robust, repeatable, accurate and fast segmentation39.                               (a)                                (b)                                                                                           (c)                                 (d)                                  (e)               Fig. 1 MRI multisequenceand tumor substructures (a) T1-weighted (b) T1c (c) FLAIR (d) T2-weighted (e)Tumor with substructuresThe abnormal image selection is a pre-processof tumor segmentation from the stack of 2D slice. Therefore efficient algorithmfor tumor detection is always demand in the field of classification which helpsto concentrate the diagnosis process with abnormal slice alone 40. In recentyears, feature extraction with support vector machine (SVM), artificial neuralnetwork (ANN), k-nearest neighbor (KNN), random forest (RF), self-organizingmaps (SOM) and deep leaning neural network (DNN) classifiers and its variantshas been proposed for MR brain tumor detection 41 – 44, 56. Fully automaticmethods for segmenting brain tumor from abnormal image are very helpful in the clinicalenvironment. In past few decades, medical imaging with following soft computingtechniques plays a important role in brain tumor segmentation to archive highaccuracy: Convolutional neural network (CNN), Local ternary pattern (LTP),Fuzzy computing and genetic algorithm4547 4950.

Segmentationalgorithms need more computational power for large dataset to produce fast andaccurate results 51. Graphics processing unit (GPU) gives the solutions tothese computation problems for large datasets using their features such as,high computation throughput, high memory bandwidth, support for floating-pointarithmetic and low cost. Compute unified device architecture (CUDA) is a GPUprogramming model introduced by NVIDIA Corporation to solve large computationalproblems 55. GPU based brain tumor segmentation methods have proposed inrecent years to achieve high computational throughput 52 -54.

In this paper, proposed a fully automaticbrain tumor detection and segmentation method using multimodal Brain TumorSegmentation (BraTS) training datasets. We categories proposed method intothree phases: (1) abnormality detection, (2) approximate tumor extraction and(3) tumor substructures segmentation. For the pre-processing, abnormal sliceswere efficiently detected from dataset using feature blocks and SVM classifier.Approximate tumor region reached using fuzzy c means (FCM) algorithm. A novel probabilisticternary pattern (PLTP) is proposed for tumor substructures segmentation fromapproximate tumor region. For the post processing, segmented substructures usedto estimate the tumor volume and 3D visualization.

Finally, availability of GPU-CUDAmakes the inherent parallelism of FCM algorithm in the proposed method toreduce the computation time. We report promising results obtained using BraTS2013 and BraTS 2015 training dataset.The rest of this paper is structures asfollows: in Section 2 describe the available prior work related with thisexperiment; Section 3 explained materials and methods used for the experiment;Section 4 illustrate the implementation details of proposed methodology; evaluation parameters are discussed in Section 5;Results and analysis are given in the Section 6; Section 7 concluded the work. 

x

Hi!
I'm Mary!

Would you like to get a custom essay? How about receiving a customized one?

Check it out