Edge preserving de-noising method for efficient segmentation of cochlear nerve by magnetic resonance imaging

Abstract

This article presents a de-noising method to improve the visual quality, edge preservation, and segmentation of cochlear nerve (CN) from magnetic resonance (MR) images. The de-noising method is based on non-local means (NLM) filter combining with stationary wavelet transform (SWT). The edge information is extracted from the residue of the NLM filter by processing it through the cycle spinning (CS). The visual interpretation of the proposed approach shows that it not only preserves CN edges but, also reduces the Gibbs phenomenon at their edges. The de-noising abilities of the proposed method strategy are assessed utilising parameters such as root mean square error (RMSE), signal-to-noise ratio (SNR), image quality index (IQI) and feature similarity index (FSIM). The efficiencies of the proposed methods are further illustrated by segmenting the cochlear nerve (CN) of the inner ear by the region growing technique. The segmentation efficiencies are evaluated by calculating the cross-sectional area (CSA) of the CN for different de-noising methods. The comparative results show the significant improvement in edge preservation of CN from MR images after de-noising the image with proposed technique.