Nikita Orlov

WND-CHARM: Multi-purpose image classification using compound image transforms

We describe a multi-purpose image classifier that can be applied to a wide variety of image classification tasks without modifications or fine-tuning, and yet provide classification accuracy comparable to state-of-the-art task-specific image classifiers. The proposed image classifier first extracts a large set of 1025 image features including polynomial decompositions, high contrast features, pixel statistics, and textures. These features are computed on the raw image, transforms of the image, and transforms of transforms of the image. The feature values are then used to classify test images into a set of pre-defined image classes. This classifier was tested on several different problems including biological image classification and face recognition. Although we cannot make a claim of universality, our experimental results show that this classifier performs as well or better than classifiers developed specifically for these image classification tasks. Our classifiers high performance on a variety of classification problems is attributed to (i) a large set of features extracted from images; and (ii) an effective feature selection and weighting algorithm sensitive to specific image classification problems. The algorithms are available for free download from openmicroscopy.org.

Wndchrm – an open source utility for biological image analysis

BackgroundBiological imaging is an emerging field, covering a wide range of applications in biological and clinical research. However, while machinery for automated experimenting and data acquisition has been developing rapidly in the past years, automated image analysis often introduces a bottleneck in high content screening.MethodsWndchrm is an open source utility for biological image analysis. The software works by first extracting image content descriptors from the raw image, image transforms, and compound image transforms. Then, the most informative features are selected, and the feature vector of each image is used for classification and similarity measurement.ResultsWndchrm has been tested using several publicly available biological datasets, and provided results which are favorably comparable to the performance of task-specific algorithms developed for these datasets. The simple user interface allows researchers who are not knowledgeable in computer vision methods and have no background in computer programming to apply image analysis to their data.ConclusionWe suggest that wndchrm can be effectively used for a wide range of biological image analysis tasks. Using wndchrm can allow scientists to perform automated biological image analysis while avoiding the costly challenge of implementing computer vision and pattern recognition algorithms.

Pattern Recognition Software and Techniques for Biological Image Analysis

The increasing prevalence of automated image acquisition systems is enabling new types of microscopy experiments that generate large image datasets. However, there is a perceived lack of robust image analysis systems required to process these diverse datasets. Most automated image analysis systems are tailored for specific types of microscopy, contrast methods, probes, and even cell types. This imposes significant constraints on experimental design, limiting their application to the narrow set of imaging methods for which they were designed. One of the approaches to address these limitations is pattern recognition, which was originally developed for remote sensing, and is increasingly being applied to the biology domain. This approach relies on training a computer to recognize patterns in images rather than developing algorithms or tuning parameters for specific image processing tasks. The generality of this approach promises to enable data mining in extensive image repositories, and provide objective and quantitative imaging assays for routine use. Here, we provide a brief overview of the technologies behind pattern recognition and its use in computer vision for biological and biomedical imaging. We list available software tools that can be used by biologists and suggest practical experimental considerations to make the best use of pattern recognition techniques for imaging assays.

IICBU 2008: a proposed benchmark suite for biological image analysis

New technology for automated biological image acquisition has introduced the need for effective biological image analysis methods. These algorithms are constantly being developed by pattern recognition and machine vision experts, who tailor general computer vision techniques to the specific needs of biological imaging. However, computer scientists do not always have access to biological image datasets that can be used for computer vision research, and biologist collaborators who can assist in defining the biological questions are not always available. Here, we propose a publicly available benchmark suite of biological image datasets that can be used by machine vision experts for developing and evaluating biological image analysis methods. The suite represents a set of practical real-life imaging problems in biology, and offers examples of organelles, cells and tissues, imaged at different magnifications and different contrast techniques. All datasets are available for free download at http://ome.grc.nia.nih.gov/iicbu2008.

Knee X-Ray Image Analysis Method for Automated Detection of Osteoarthritis

We describe a method for automated detection of radiographic osteoarthritis (OA) in knee X-ray images. The detection is based on the Kellgren-Lawrence (KL) classification grades, which correspond to the different stages of OA severity. The classifier was built using manually classified X-rays, representing the first four KL grades ( normal, doubtful, minimal, and moderate). Image analysis is performed by first identifying a set of image content descriptors and image transforms that are informative for the detection of OA in the X-rays and assigning weights to these image features using Fisher scores. Then, a simple weighted nearest neighbor rule is used in order to predict the KL grade to which a given test X-ray sample belongs. The dataset used in the experiment contained 350 X-ray images classified manually by their KL grades. Experimental results show that moderate OA (KL grade 3) and minimal OA (KL grade 2) can be differentiated from normal cases with accuracy of 91.5% and 80.4%, respectively. Doubtful OA (KL grade 1) was detected automatically with a much lower accuracy of 57%. The source code developed and used in this study is available for free download at www.openmicroscopy.org.

Impressionism, expressionism, surrealism: Automated recognition of painters and schools of art

We describe a method for automated recognition of painters and schools of art based on their signature styles and studied the computer-based perception of visual art. Paintings of nine artists, representing three different schools of art—impressionism, surrealism and abstract expressionism—were analyzed using a large set of image features and image transforms. The computed image descriptors were assessed using Fisher scores, and the most informative features were used for the classification and similarity measurements of paintings, painters, and schools of art. Experimental results show that the classification accuracy when classifying paintings into nine painter classes is 77%, and the accuracy of associating a given painting with its school of art is 91%. An interesting feature of the proposed method is its ability to automatically associate different artists that share the same school of art in an unsupervised fashion. The source code used for the image classification and image similarity described in this article is available for free download.

Computer Vision for Microscopy Applications

The tremendous growth in digital imagery has introduced the need for accurate image analysis and classification. The applications include content based image retrieval in the World Wide Web and digital libraries (Dong & Yang, 2002; Heidmann, 2005; Smeulders et al., 2000; Veltkamp et al., 2001) scene classification (Huang et al., 2005; Jiebo et al., 2005), face recognition (Jing & Zhang, 2006; Pentland & Choudhury, 2000; Shen & Bai, 2006) and biological and medical image classification (Awate et al., 2006; Boland & Murphy, 2001; Cocosco et al., 2004; Ranzato et al., 2007). Although attracting considerable attention in the past few years, image classification is still considered a challenging problem in machine learning due to the complexity of real-life images. This chapter discusses an approach to computer vision using automated image classification and similarity measurement based on a large set of general image descriptors. Classification results as well as image similarity measurements are presented for several diverse applications.

Automatic Classification of Lymphoma Images With Transform-Based Global Features

We propose a report on automatic classification of three common types of malignant lymphoma: chronic lymphocytic leukemia, follicular lymphoma, and mantle cell lymphoma. The goal was to find patterns indicative of lymphoma malignancies and allowing classifying these malignancies by type. We used a computer vision approach for quantitative characterization of image content. A unique two-stage approach was employed in this study. At the outer level, raw pixels were transformed with a set of transforms into spectral planes. Simple (Fourier, Chebyshev, and wavelets) and compound transforms (Chebyshev of Fourier and wavelets of Fourier) were computed. Raw pixels and spectral planes were then routed to the second stage (the inner level). At the inner level, the set of multipurpose global features was computed on each spectral plane by the same feature bank. All computed features were fused into a single feature vector. The specimens were stained with hematoxylin (H) and eosin (E) stains. Several color spaces were used: RGB, gray, CIE-L*a*b*, and also the specific stain-attributed H&E space, and experiments on image classification were carried out for these sets. The best signal (98%-99% on earlier unseen images) was found for the HE, H, and E channels of the H&E data set.

Pattern recognition approaches to compute image similarities: application to age related morphological change

We are studying the genetic influence on rates of age related muscle degeneration in C. elegans. For this, we built pattern recognition tools to calculate a morphological score given an image of muscle tissue. We collected images of body wall muscle and the terminal bulb of the pharynx at four different ages. We extracted a large set of image descriptors (signatures) from both sets of images. Two different methods were used for pattern recognition within these two datasets. Both methods compute a single number that correlates with the known age of the sample. Because aging is a continuous process, the relative age computed from images of tissue can be viewed as a measure of image similarity. The techniques employed and validated in this work can be generalized to other areas such as image-based queries

Molecular characterization of the transition to mid-life in Caenorhabditis elegans.

We present an initial molecular characterization of a morphological transition between two early aging states. In previous work, an age score reflecting physiological age was developed using a machine classifier trained on images of worm populations at fixed chronological ages throughout their lifespan. The distribution of age scores identified three stable post-developmental states and transitions. The first transition occurs at day 5 post-hatching, where a significant percentage of the population exists in both state I and state II. The temperature dependence of the timing of this transition (Q10 ~ 1.17) is too low to be explained by a stepwise process with an enzymatic or chemical rate-limiting step, potentially implicating a more complex mechanism. Individual animals at day 5 were sorted into state I and state II groups using the machine classifier and analyzed by microarray expression profiling. Despite being isogenic, grown for the same amount of time, and indistinguishable by eye, these two morphological states were confirmed to be molecularly distinct by hierarchical clustering and principal component analysis of the microarray results. These molecular differences suggest that pharynx morphology reflects the aging state of the whole organism. Our expression profiling yielded a gene set that showed significant overlap with those from three previous age-related studies and identified several genes not previously implicated in aging. A highly represented group of genes unique to this study is involved in targeted ubiquitin-mediated proteolysis, including Skp1-related (SKR), F-box-containing, and BTB motif adaptors.