Improved Segmentation of MRI Brain Images by Denoising and Contrast Enhancement

Abstract

Background/Objectives: The Rician noise in MRI (Magnetic Resonance Image) degrades the image quality and thus, accuracy in segmentation is reduced and localization of tumour may not be precise. In this paper, a robust approach is proposed which estimates and removes the Rician noise of 2D MRI for improving segmentation and detection of tumours. Methods/Statistical analysis: First, a robust Rician noise estimation algorithm is employed to identify all the pixels with high Rician noise. Second, a bilateral filter based denoising algorithm is employed to filter image in the wavelet domain. Successively a bilateral filter parameter optimization method is adopted, which uses the noise, contrast and frequency components in MRI to select suitable filter parameters for Bilateral Filter (BF). It is suitable for edge preserving and for adaptive denoising to segment image correctly. Further, after denoising the image, the contrast of the image is improved as a pre-processing step before the image segmentation. Next, SVM-based image segmentation algorithm is employed to segment the 2D MRI. Findings: The algorithm is tested both in synthetic and real-time clinical images of tumour affected human brain. The simulation tests show that the denoising and contrast enhancement improves the segmentation of images. The performance of the proposed approach is improved by 29% in segmentation of synthetic images compared to the existing similar techniques. Similarly, an improvement of 22% in segmentation is observed for real-time images. Application/Improvements: This approach shows comparable improvement in with respect to processing of MRI. The same procedure may be adopted for other imaging techniques.