Noise reduction filter optimization for prostate cancer localization

Abstract

Multispectral Magnetic Resonance Imaging (MRI) images are commonly used in prostate cancer diagnosis. However noise in raw data makes it difficult to process images. Therefore MR images must be filtered as a pre-processing step prior to automated localization. In the literature, filters and their parameters are generally selected depending on experiences in the field. In this research, a method, not found in the literature, is proposed such that the system can choose optimal filter parameters to maximize cancer localization. In order to use on KEL, KEP, and IAUC 30, 60, 90 parameters, obtained from multispectral MR images, 3 different filters (wiener filter, total variance filter and wavelet thresholding) and a parameter for each filtering is chosen to maximize localization performance. Linear discriminant analysis is used for localization and observed that optimally selecting the filtering method and its parameter improves prostate cancer localization performance. © 2019 IEEE.