A Compact Representation of Histopathology Images Using Digital Stain Separation and Frequency-Based Encoded Local Projections

Abstract

In recent years, histopathology images have been increasingly used as a diagnostic tool in the medical field. The process of accurately diagnosing a biopsy sample requires significant expertise in the field, and as such can be time-consuming and is prone to uncertainty and error. With the advent of digital pathology, using image recognition systems to highlight problem areas or locate similar images can aid pathologists in making quick and accurate diagnoses. In this paper, we specifically consider the encoded local projections (ELP) algorithm, which has previously shown some success as a tool for classification and recognition of histopathology images. We build on the success of the ELP algorithm as a means for image classification and recognition by proposing a modified algorithm which captures the local frequency information of the image. The proposed algorithm estimates local frequencies by quantifying the changes in multiple projections in local windows of greyscale images. By doing so we remove the need to store the full projections, thus significantly reducing the histogram size, and decreasing computation time for image retrieval and classification tasks. Furthermore, we investigate the effectiveness of applying our method to histopathology images which have been digitally separated into their hematoxylin and eosin stain components. The proposed algorithm is tested on the publicly available invasive ductal carcinoma (IDC) data set. The histograms are used to train an SVM to classify the data. The experiments showed that the proposed method outperforms the original ELP algorithm in image retrieval tasks. On classification tasks, the results are found to be comparable to state-of-the-art deep learning methods and better than many handcrafted features from the literature.

Studying the Effect of Digital Stain Separation of Histopathology Images on Image Search Performance

“…For each descriptor, and each set of input colour channels, the best accuracy, taken over all distance functions, is presented. It should be noted here that there are some slight discrepancies between the results presented here for the IDC data set and those in our previous work [4]. This is due to a small error which was found in the code which slightly changes the numerical results, but does not change the overall conclusions of the previous study.…”

“…In this paper, as in [4], we adopt the stain separation method proposed in [12], an extension of the wedge finding method from [10]. Unlike some previous methods for stain separation [16], this method does not require any calibration or knowledge of the exact stain colours, instead it works by using the available image data to estimate an H&E basis.…”

“…In our study, we implement four image descriptors, including the LBP descriptor, along with the Gabor filter-based GIST descriptor [15] which computes the spatial envelope of a scene, the ELP descriptor (encoded local projections) from [18] which was designed with medical images in mind, and our proposed F-ELP descriptor from [4], designed specifically to be a compact descriptor for histopathology images.…”

“…In order to properly use the relevant colour information, a WSI analysis system should be able to process colour images and determine relevant biological information from the presence of different stain colours. In a previous work [4], we introduced a new frequency-based ELP (F-ELP) image descriptor for histopathology image search, which captures local frequency information and implemented this new descriptor on images which were separated into two colour channels based on chemical stain components using a digital stain separation algorithm. In [4], we found that both the ELP and F-ELP descriptors saw improved search accuracy when applied to the stain-separated images, as opposed to single-channel greyscale images.…”

“…In a previous work [4], we introduced a new frequency-based ELP (F-ELP) image descriptor for histopathology image search, which captures local frequency information and implemented this new descriptor on images which were separated into two colour channels based on chemical stain components using a digital stain separation algorithm. In [4], we found that both the ELP and F-ELP descriptors saw improved search accuracy when applied to the stain-separated images, as opposed to single-channel greyscale images. In this paper, we focus on studying the effectiveness of digital stain separation of histopathology images for image retrieval applications using a number of common handcrafted image descriptors.…”