Visual importance pooling for image quality assessment of despeckle filters in Optical Coherence Tomography

Abstract

Recent methods of image quality assessment emphasise on pooling of multiple criteria of visual importance for comparing across different natural images. An important applicational aspect of pooled assessment is in selection of a desired image processing operator from a family of available such. Medical imaging has matured of ages utilizing multiple methods of image processing to achieve similar objectives of feature specific image augmentation, while lacking frameworks for comparatively assessing their performance. In this work we present such a framework using three visual importance pooling criteria based image quality assessment metrics for comparative evaluation of seven types of despeckle filters applied to Optical Coherence Tomography (OCT). Along with this evaluation, conventional methods of image quality assessment including Mean Squared Error (MSE), Minkowski metric (∈β), and Singal-to-Noise Ratio (SNR) are also used. A note about the impact of the used despeckle filters on visual perception of the expert examining the OCT images is also provided along with a correlated inference on image quality as assessed using the pooling criteria. The results of this study suggest that geometric filtering algorithm outperforms other candidate methods with best representation of universal image quality, mean structural similarity, and mean gradient-based structural similarity. Also this speckle reduction method is associated with SNR > 50dB, and mimimum of MSE and ∈β. This work on a larger scale demonstrate a useful framework for comparative assessment of similar image quality improvement algorithms.