Computer Aided Diagnosis System to Identify the Cochlear Nerve from Magnetic Resonance MR Images

Abstract

Sensorineural Hearing Loss (SNHL) is a hearing impairment resulting from a pathological condition in the inner ear or along the auditory nerve (VIII cranial nerve) pathway to the brainstem. It is critical to the diagnosis and treats the children with the congenital hearing loss for their early speech and language development, academic achievement, social and emotional development. Cochlear Implant (CI) is an effective treatment option for pediatric patients with congenital SNHL. The size of the cochlear nerve (CN) precludes the successful outcome of cochlear implant surgery. Traditional measurement of cochlear nerve involves the expert radiologist to segment and measure the cochlear nerve manually. However, because of the enormous computation of the medical image information, this manual measurement is much tiresome and can be exceedingly subjective due to inter-observer variability. Motivated by advanced computer technology, Computer-Aided Diagnosis (CAD) newlinecan help the radiologists in estimating the cochlear nerve size. The CAD system presented in this thesis can be utilized to identify and measure the cochlear nerve size of the magnetic resonance images. The main objective of the thesis is to establish new algorithms for pre-processing, segmentation, measurement and classification of the cochlear nerve from the inner ear MR images. newlineAlthough the acquisition speed and resolution of the MR images are improved, they newlineare affected by noise. Since the noise in MR images is a signal dependent Rician noise,the reduction of noise is still a difficult task. In the first stage of the CAD system, the Rician noise in the Magnetic resonance image is removed by proposed Non-Local Means based Stationary Wavelet Transform (NLSWT) method. Later on, this method is extended to sharpen the subtle structure edges and to reduce the Gibb s phenomenon using the proposed Laplacian Pyramid based Singular Value Decomposition (LPSVD) method. The experimental result shows that the pre-processing greatly improves the visual quality and edge sh