Brain Tumor Segmentation using Watershed Technique and Self Organizing Maps

Ashima Anand

doi:10.17485/ijst/2017/v10i44/120574

Abstract

Objectives: To segment tumor with higher accuracy. Methods/Statistical Analysis: Noise removal is done with the help of Gabor filter as a preprocessing step. Skull stripping is done to remove non cerebral regions using thresholding and morphological operations. Segmentation using watershed algorithm is done, as it achieves exact location of outline. Unsupervised type of neural network i.e. self organizing maps is used for classification. Finding: It has been analyzed that by combining watershed and neural networks segmentation accuracy has been improved to 95.93%. The motive of the research is to segment the tumor with precision using computerized segmentation algorithm that can help physicians to analyze brain diseases and treatment can be started as soon as possible. Applications: The proposed technique can be used in image processing of brain tumor detection. Keywords: Brain Tumor Segmentation, Image Segmentation, Magnetic Resonance Imaging (MRI), Self Organizing Maps (SOM), Stationary Wavelet Transform (SWT)

Highlights

Tumors may be grouped into primary and secondary[1]
Brain tumor is detected by medical examination through various imaging modalities such as CAT and MRI2
Segmentation involves the process of splitting up the image into distinct regions i.e. according to criteria of homogeneity[3]

Summary

Introduction

Tumors may be grouped into primary and secondary[1]. Primary or brain metastasis tumors may initiate in the brain or membranes, nerves or glands which is further categorized into Benign (not causes cancer) and Malignant (prone to cancer). Malignant brain tumor is characterized as threatful, which invade rapidly, destroying brain cells by causing swelling. The exact cause of brain tumors is not clear. Brain tumor is detected by medical examination through various imaging modalities such as CAT and MRI2. Segmentation of brain tumor is one of the competitive tasks since tumor’s characteristics are very difficult to visualize[4]. Various challenges related to tumor segmentation are high diversity appearance and inconsistent shape. Segmentation of tumor done in a manual manner by doctors is a weary task which shows variations when diverse doctors undergo the same task of segmentation

Objectives

Methods

Results

Conclusion