Texture analysis of 3D fluorescence microscopy images using RSurf 3D features

Classification tasks of biomedical images are still an interesting topic of research with many possibilities of improvement. A very important part in these tasks is the feature extraction, where different image descriptors are used. Recently, a new approach of RSurf features was introduced with application in recognition of the 2D HEp-2 cell images. In this work, we present the extension of these features for the 3D volumetric images and demonstrate its superiority in recognition of sub-cellular protein distribution. The performance is tested on public HeLa dataset containing 9 unique image classes. The k-NN classifier based purely on the RSurf 3D features achieves more than 99% accuracy in recognition of the 3D HeLa images.