A Multiresolution Enhancement to Generic Classifiers of Subcellular Protein Location Images

Abstract

We propose an algorithm for the classification of fluorescence microscopy images depicting the spatial distribution of proteins within the cell. The problem is at the forefront of the current trend in biology towards understanding the role and function of all proteins. The importance of protein subcellular location was pointed out by Murphy, whose group produced the first automated system for classification of images depicting these locations, based on diverse feature sets and combinations of classifiers. With the addition of the simplest multiresolution features, the same group obtained the highest reported accuracy of 91.5% for the denoised 2D HeLa data set. Here, we aim to improve upon that system by adding the true power of multiresolution-adaptivity. In the process, we build a system able to work with any feature sets and any classifiers, which we denote as a Generic Classification System (GCS). Our system consists of multiresolution (MR) decomposition in the front, followed by feature computation and classification in each subband, yielding local decisions. This is followed by the crucial step of combining all those local decisions into a global one, while at the same time ensuring that the resulting system does no worse than a no-decomposition one. On a nondenoised data set and a much smaller number of features (a combination of texture and Zernicke moment features) and a neural network classifier, we obtain a high accuracy of 89.8%, effectively proving that the space-frequency localized information in the subbands adds to the discriminative power of the system.

A multiresolution approach to automated classification of protein subcellular location images

“…The challenge in this data set is that images from the same class may look different while those from different classes may look very similar (see Figure 2 in [13]). Based on the above discussion, we would like to extract discriminative features within space-frequency localized subspaces.…”

“…The open-form algorithm for the training and the testing phases are given in [13] under Algorithms 1 and 2, respectively.…”

“…We now summarize our initial MR classification effort [12,13]. We started with a simple classification system consisting of Haralick texture feature computation followed by a maximum-likelihood classifier, and demonstrated that, by adding an MR block in front, we were able to raise the classification accuracy by roughly 10% (from 71.8% to 82.2%) as compared to the system with no MR.…”

Towards an Image Analysis Toolbox for High-Throughput Drosophila Embryo Rnai Screens

“…In our previous work [11] [16], we found that texture features are the most discriminative. We modified the standard Haralick texture features we call T1 [17] into a new set T3, by separating vertical/horizontal from diagonal features, as follows:…”

“…In [11], we introduced a concept of multiresolution (MR) classification for classification of protein subcellular location images, arguing that the nature of such images requires tools which offer localization in space and frequency as well as adaptivity. Thus, we classify in MR subspaces as opposed on the original image itself with the idea is that certain features will react well at a certain scale but not at another.…”