Feature extraction from cancer images using local phase congruency: A reliable source of image descriptors

Abstract

The (semi-) automatic detection of significant (feature) points is a key task in in vivo assessment of cancer staging and progression. However, this is a challenging task due to the relatively poor signal-to-noise, limited resolution and variable intensity of medical images. We propose to use phase congruence (PC), the Morrone and Owens (1987) feature model, to extract local descriptors. We overcome the limitations of the currently accepted PC measure, estimate PC without using an image energy weighting factor. We show that: (i) relative phase values from a single scale are not equivalent to phase values from PC, and should not be used to assess local image structure; and (ii) our approach results in higher specificity to features of interest, and lower sensitivity to noise, as demonstrated in in vitro microscopy (e.g. tumour microvessels) and in vivo pre-clinical pancreatic cancer images.