Feature normalization via expectation maximization and unsupervised nonparametric classification for M-FISH chromosome images

Multicolor fluorescence in situ hybridization (M-FISH) techniques provide color karyotyping that allows simultaneous analysis of numerical and structural abnormalities of whole human chromosomes. Chromosomes are stained combinatorially in M-FISH. By analyzing the intensity combinations of each pixel, all chromosome pixels in an image are classified. Often, the intensity distributions between different images are found to be considerably different and the difference becomes the source of misclassifications of the pixels. Improved pixel classification accuracy is the most important task to ensure the success of the M-FISH technique. In this paper, we introduce a new feature normalization method for M-FISH images that reduces the difference in the feature distributions among different images using the expectation maximization (EM) algorithm. We also introduce a new unsupervised, nonparametric classification method for M-FISH images. The performance of the classifier is as accurate as the maximum-likelihood classifier, whose accuracy also significantly improved after the EM normalization. We would expect that any classifier will likely produce an improved classification accuracy following the EM normalization. Since the developed classification method does not require training data, it is highly convenient when ground truth does not exist. A significant improvement was achieved on the pixel classification accuracy after the new feature normalization. Indeed, the overall pixel classification accuracy improved by 20% after EM normalization.     

Chromenet: a CNN architecture with comparison of optimizers for classification of human chromosome images

Multidimensional Systems and Signal Processing - Chromosome karyotyping is pivotal in the diagnosis of many genetic disorders and birth defects. Manual karyotyping is a tedious procedure and...     

Efficient classification of ISAR images

In this paper, we propose a method to classify inverse synthetic aperture radar images from different targets. Our approach can provide efficient features for classification by the combined use of a polar mapping procedure and a well-designed classifier. The resulting feature vectors are able to meet the requirements that efficient features should have: invariance with respect to rotation and scale, small dimensionality, as well as highly discriminative information. Typical experimental examples of the proposed method are provided and discussed.     

Model-based classification of radar images

A Bayesian approach is presented for model-based classification of images with application to synthetic-aperture radar. Posterior probabilities are computed for candidate hypotheses using physical features estimated from sensor data along with features predicted from these hypotheses. The likelihood scoring allows propagation of uncertainty arising in both the sensor data and object models. The Bayesian classification, including the determination of a correspondence between unordered random features, is shown to be tractable, yielding a classification algorithm, a method for estimating error rates, and a tool for evaluating the performance sensitivity. The radar image features used for classification are point locations with an associated vector of physical attributes; the attributed features are adopted from a parametric model of high-frequency radar scattering. With the emergence of wideband sensor technology, these physical features expand interpretation of radar imagery to access the frequency- and aspect-dependent scattering information carried in the image phase     

Subspace-based and DIRECT algorithms for distorted circular contour estimation.

Circular features are commonly sought in digital image processing. The subspace-based line detection (SLIDE) method proposed to estimate the center and the radius of a single circle. In this paper, we introduce a novel method for estimating several radii while extending the circle estimation to retrieve circular-like distorted contours. Particularly, we develop and validate a new model for virtual signal generation by simulating a circular antenna. The circle center is estimated by the SLIDE method. A variable speed propagation scheme toward the circular antenna yields a linear phase signal. Therefore, a high-resolution method provides the radius. Either the gradient method or the more robust combination of dividing rectangles and spline interpolation can extend this method extend this method for free form object segmentation. The retrieval of multiple non concentric circles and rotated ellipses is also considered. To evaluate the performance of the proposed methods, we compare them with a least-squares method, Hough transform, and gradient vector flow. We apply the proposed method to hand-made images while considering some real-world images.     

Cascaded differential and wavelet compression of chromosome images

This paper proposes a new method for chromosome image compression based on an important characteristic of these images: the regions of interest (ROIs) to cytogeneticists for evaluation and diagnosis are well determined and segmented. Such information is utilized to advantage in our compression algorithm, which combines lossless compression of chromosome ROIs with lossy-to-lossless coding of the remaining image parts. This is accomplished by first performing a differential operation on chromosome ROIs for decorrelation, followed by critically sampled integer wavelet transforms on these regions and the remaining image parts. The well-known set partitioning in hierarchical trees (SPIHT) (Said and Perlman, 1996) algorithm is modified to generate separate embedded bit streams for both chromosome ROIs and the rest of the image that allow continuous lossy-to-lossless compression of both (although lossless compression of the former is commonly used in practice). Experiments on two sets of sample chromosome spread and karyotype images indicate that the proposed approach significantly outperforms current compression techniques used in commercial karyotyping systems and JPEG-2000 compression, which does not provide the desirable support for lossless compression of arbitrary ROIs.     

Bayesian fused classification of medical images

In many applications in computer vision and signal processing, it is necessary to assimilate data from multiple sources. This is a particularly important issue in medical imaging, where information on a patient may be available from a number of different modalities. As a result, there has been much recent research interest in this area. The authors suggest an additional Bayesian method which generates a segmented classification concurrently with improving reconstructions of a set of registered images. A synthetic example is used to demonstrate the subjectives and benefits of this proposed approach. Two medical applications, one fusing computed tomography (CT) and single photon emission computed tomography (SPECT) brain scans, and the other magnetic resonance (MR) images at two different resolutions, are considered.     

Knowledge-based classification of polarimetric SAR images

In preparation for the flight of the Shuttle Imaging Radar-C (SIR-C) on board the Space Shuttle in the spring of 1994, a level-1 automatic classifier was developed on the basis of polarimetric SAR images acquired by the JPL AirSAR system. The classifier uses L- and C-Band polarimetric SAR measurements of the imaged scene to classify individual pixels into one of four categories: tall vegetation (trees), short vegetation, urban, or bare surface, with the last category encompassing water surfaces, bare soil surfaces, and concrete or asphalt-covered surfaces. The classifier design uses knowledge of the nature of radar backscattering from surfaces and volumes to construct appropriate discriminators in a sequential format. The classifier, which was developed using training areas in a test site in Northern Michigan, was tested against independent test areas in the same test site and in another site imaged three months earlier. Among all cases and all categories, the classification accuracy ranged between 91% and 100%     

Scene-oriented hierarchical classification of blurry and noisy images

A system for scene-oriented hierarchical classification of blurry and noisy images is proposed. It attempts to simulate important features of the human visual perception. The underlying approach is based on three strategies: extraction of essential signatures captured from a global context, simulating the global pathway; highlight detection based on local conspicuous features of the reconstructed image, simulating the local pathway; and hierarchical classification of extracted features using probabilistic techniques. The techniques involved in hierarchical classification use input from both the local and global pathways. Visual context is exploited by a combination of Gabor filtering with the principal component analysis. In parallel, a pseudo-restoration process is applied together with an affine invariant approach to improve the accuracy in the detection of local conspicuous features. Subsequently, the local conspicuous features and the global essential signature are combined and clustered by a Monte Carlo approach. Finally, clustered features are fed to a self-organizing tree algorithm to generate the final hierarchical classification results. Selected representative results of a comprehensive experimental evaluation validate the proposed system.     

Detection and classification of edges in color images

Up to now, most of the color edge detection methods are monochromatic-based techniques, which produce, in general, better than when traditional gray-value techniques are applied. In this overview, we focus mainly on vector-valued techniques because it is easy to understand how to apply common edge detection schemes to every color component. Opposed to this, vector-valued techniques are new and different. The second part of the article addresses the topic of edge classification. While edges are often classified into step edges and ramp edges, we address the topic of physical edge classification based on their origin into shadow edges, reflectance edges, orientation edges, occlusion edges, and specular edges. In the rest of this article we discuss various vector-valued techniques for detecting discontinuities in color images. Then operators are presented based on vector order statistics, followed by presentation by examples of a couple of results of color edge detection. We then discuss different approaches to a physical classification of edges by their origin.     

Fuzzy supervised classification of remote sensing images

A fuzzy supervised classification method in which geographical information is represented as fuzzy sets is described. The algorithm consists of two major steps: the estimate of fuzzy parameters from fuzzy training data, and a fuzzy partition of spectral space. Partial membership of pixels allows component cover classes of mixed pixels to be identified and more accurate statistical parameters to be generated, resulting in a higher classification accuracy. Results of classifying a Landsat MSS image are presented, and their accuracy is analyzed     

Modelling and classification of vascular smooth muscle cell images

A technique integrating directional fractals and neural networks for modelling and classification of tissue section images of smooth muscle cells in human resistance arteries is presented. The ability of directional fractals in capturing the orientation and roughness degrees of texture images is shown. This ability entails improved results of tissue classification over the use of isotropic multiresolution fractals     

Detection and classification of circular structures on SPOT images

An unsupervised method for detecting structural features on satellite images that consists of three major steps is proposed. The extraction of contours, which depends on the texture encountered is achieved by an iterative filtering followed by several thresholds that generate binary images. The detection of the structures from the contours involves four substeps: individualization of the curves; decomposition of the curves into subcircular elements; application of a version of the Hough transform to each subcircular element; and computation of precise results. The parameters that discriminate the structures are computed. These data allow the selection of different families of structures that are being looked at. The method has been applied to the region of Azru     

Subspace-based estimation of time delays and Doppler shifts

This paper considers the problem of estimating the time delays and Doppler shifts of a known waveform received via several distinct paths by an array of antennas. The general maximum likelihood estimator is presented, and is shown to require a 2d-dimensional nonlinear minimization, where d is the number of received signal reflections. Two alternative solutions based on signal and noise subspace fitting are proposed, requiring only a d-dimensional minimization. In particular, we show how to decouple the required search into a two-step procedure, where the delays are estimated and the Dopplers solved for explicitly. Initial conditions for the time delay search can be obtained by applying generalizations of the MUSIC and ESPRIT algorithms, which are also outlined in the paper. Simulation examples are included to illustrate the algorithms' performance relative to the Cramer-Rao bound     

Fourier classification images in cardiac nuclear medicine

At present, the diagnosis of cardiac left ventricular regional wall motion abnormalities (RWMA) in nuclear medicine is aided mainly by phase images and amplitude images, which picture the spatial distribution of the phase and of the amplitude of the first harmonics of pixel time activity curves, respectively. However, they do not utilize other information contained in the original radionuclide images, and they do not offer a direct diagnostic interpretation of the data. The proposed Fourier classification images (FCI) overcome these deficiencies. Their pixel intensities express directly the diagnostic class of RWMA. The FCI pixel intensities are functions of pixel coordinates, Fourier features of pixel time activity curves, and their distribution parameters, and they are not limited by the first harmonics model. The derivation of the pixel classifier includes normalization transformation of coordinates and activities. Fourier analysis of raw image data, and teaching the computer by examples of already diagnosed cases with the help of discriminant analysis. FCI offer direct and robust diagnosis of RWMA, superior to that derived from phase and amplitude images, especially in the detection of mild RWMA.     

Linear structures in mammographic images: detection and classification

We describe methods for detecting linear structures in mammograms, and for classifying them into anatomical types (vessels, spicules, ducts, etc). Several different detection methods are compared, using realistic synthetic images and receiver operating characteristic (ROC) analysis. There are significant differences (p < 0.001) between the methods, with the best giving an Az value for pixel-level detection of 0.943. We also investigate methods for classifying the detected linear structures into anatomical types, using their cross-sectional profiles, with particular emphasis on recognising the "spicules" and "ducts" associated with some of the more subtle abnormalities. Automatic classification results are compared with expert annotations using ROC analysis, demonstrating useful discrimination between anatomical classes (Az = 0.746). Some of this discrimination relies on simple attributes such as profile width and contrast, but important information is also carried by the shape of the profile (Az = 0.653). The methods presented have potentially wide application in improving the specificity of abnormality detection by exploiting additional anatomical information.     

Projection pursuit classification of multiband polarimetric SARland images

Results are presented for an experiment utilizing a pastoral land scene with a variety of eight classes, imaged by the NRL dual band (X and L) polarimetric synthetic aperture radar (NUWSAR) at a spatial resolution of 1.2 m. Projection pursuit (PP) statistical analysis tools were applied to a set of simultaneous L-band and X-band fully polarized images (six independent channels) to demonstrate the utility of land classification at high spatial resolution from a light aircraft using SAR. The statistical confusion matrix was used as a quantitative optimization measure of classification. Samples of eight classes from a portion of the scene were used to define a training set, then PP tools were used for classification. It is clear that L-band and X-band fully polarized data view the classes in a significantly different manner, and each brings independent information to the analysis. These results are not meant to be exhaustive at this time but to demonstrate the utility of applying PP tools to multiband and polarization SAR data and to give an indication of the quality of classification one can achieve with moderately high spatial resolution SAR data using a light plane platform     

Topology-preserving tissue classification of magnetic resonance brain images

This paper presents a new framework for multiple object segmentation in medical images that respects the topological properties and relationships of structures as given by a template. The technique, known as topology-preserving, anatomy-driven segmentation (TOADS), combines advantages of statistical tissue classification, topology-preserving fast marching methods, and image registration to enforce object-level relationships with little constraint over the geometry. When applied to the problem of brain segmentation, it directly provides a cortical surface with spherical topology while segmenting the main cerebral structures. Validation on simulated and real images characterises the performance of the algorithm with regard to noise, inhomogeneities, and anatomical variations.     

Unsupervised classification of cell images using pyramid nodelinking

In this communication we describe a segmentation technique which combines two properties in an iterative and hierarchial matter to correctly segment and classify the given cell images. The technique is applied to digital images taken from microscope slides of cultured rat liver cells, and the goal is to classify these cells into one of three possible classes. The first class cells (I) are morphologically normal and stain the darkest. The second class cells (II) are slightly damaged showing both nuclear and cytoplasmic swelling with resultant lessening of staining affinity. The third class cells (III) are markedly damaged as demonstrated by the presence of cytoplasmic vacuolization, or are completely disintegrated. First class cells are classified by taking advantage of their staining affinity; the original gray level image is segmented into four gray levels. The darkest is then classified as type I. Type III cells are classified by using high business as a characteristic; the standard deviation of the original image is segmented into four business levels. The highest level is classified as type III cell. Assuming only the three cell types are present in any given image, the remaining non-background unclassified pixels are determined to belong to type II cells.