Anastasios Delopoulos

Invariant image classification using triple-correlation-based neural networks

Triple-correlation-based neural networks are introduced and used in this paper for invariant classification of 2D gray scale images. Third-order correlations of an image are appropriately clustered, in spatial or spectral domain, to generate an equivalent image representation that is invariant with respect to translation, rotation, and dilation. An efficient implementation scheme is also proposed, which is robust to distortions, insensitive to additive noise, and classifies the original image using adequate neural network architectures applied directly to 2D image representations. Third-order neural networks are shown to be a specific category of triple-correlation-based networks, applied either to binary or gray-scale images. A simulation study is given, which illustrates the theoretical developments, using synthetic and real image data.

Optimal filter banks for signal reconstruction from noisy subband components

Conventional design techniques for analysis and synthesis filters in subband processing applications guarantee perfect reconstruction of the original signal from its subband components. The resulting filters, however, lose their optimality when additive noise due, for example, to signal quantization, disturbs the subband sequences. We propose filter design techniques that minimize the reconstruction mean squared error (MSE) taking into account the second order statistics of signals and noise in the case of either stochastic or deterministic signals. A novel recursive, pseudo-adaptive algorithm is proposed for efficient design of these filters. Analysis and derivations are extended to 2-D signals and filters using powerful Kronecker product notation. A prototype application of the proposed ideas in subband coding is presented. Simulations illustrate the superior performance of the proposed filter banks versus conventional perfect reconstruction filters in the presence of additive subband noise.

Designing and implementing the transition to a fully digital hospital

The increase in the number of examinations performed in modern healthcare institutions in conjunction with the range of imaging modalities available today have resulted in a tremendous increase in the number of medical images generated and has made the need for a dedicated system able to acquire, distribute, and store medical image data very attractive. Within the framework of the Hellenic R&D program, we have designed and implemented a picture archiving and communication system for a high-tech cardiosurgery hospital in Greece. The system is able to handle in a digital form images produced from ultrasound, X-ray angiography, /spl gamma/-camera, chest X-rays, as well as electrocardiogram signals. Based on the adoption of an open architecture highly relying on the DICOM standard, the system enables the smooth transition from the existing procedures to a fully digital operation mode and the integration of all existing medical equipment to the new central archiving system.

Consistent identification of stochastic linear systems with noisy input-output data

Abstract A novel criterion is introduced for parametric errors-in-variables identification of stochastic linear systems excited by non-Gaussian inputs. The new criterion is (at least theoretically) insensitive to a class of input-output disturbances because it implicitly involves higher- than second-order cumulant statistics. In addition, it is shown to be equivalent to the conventional Mean-Squared Error (MSE) as if the latter was computed in the ideal case of noise-free input-output data. The sampled version of the criterion converges to the novel MSE and guarantees strongly consistent parameter estimators. The asymptotic behavior of the resulting parameter estimators is analyzed and guidelines for minimum variance experiments are discussed briefly. Informative enough input signals and persistent of excitation conditions are specified. Computatonally attractive Recursive-Least-Squares variants are also developed for on-line implementation of ARMA modeling, and their potential is illustrated by applying them to time-delay estimation in low SNR environment. The performance of the proposed algorithms and comparisons with conventional methods are corroborated using simulated data.

Strongly consistent identification algorithms and noise insensitive MSE criteria

Windowed cumulant projections of nonGaussian linear processes yield autocorrelation estimators which are immune to additive Gaussian noise of unknown covariance. By establishing strong consistency of these estimators, strongly consistent and noise insensitive recursive algorithms are developed for parameter estimation. These computationally attractive schemes are shown to be optimal with respect to a modified mean-square-error (MSE) criterion which implicitly exploits the high signal-to-noise ratio domain of cumulant statistics. The novel MSE objective function is expressed in terms of the noisy process, but it is shown to be a scalar multiple of the standard MSE criterion as if the latter was computed in the absence of noise. Simulations illustrate the performance of the proposed algorithms and compare them with the conventional algorithms. >

Two-dimensional filter bank design for optimal reconstruction using limited subband information

In this correspondence, we propose design techniques for analysis and synthesis filters of 2-D perfect reconstruction filter banks (PRFBs) that perform optimal reconstruction when a reduced number of subband signals is used. Based on the minimization of the squared error between the original signal and some low-resolution representation of it, the 2-D filters are optimally adjusted to the statistics of the input images so that most of the signals energy is concentrated in the first few subband components. This property makes the optimal PRFBs efficient for image compression and pattern representations at lower resolutions for classification purposes. By extending recently introduced ideas from frequency domain principal component analysis to two dimensions, we present results for general 2-D discrete nonstationary and stationary second-order processes, showing that the optimal filters are nonseparable. Particular attention is paid to separable random fields, proving that only the first and last filters of the optimal PRFB are separable in this case. Simulation results that illustrate the theoretical achievements are presented.

Semantic Integration of Cervical Cancer Data Repositories to Facilitate Multicenter Association Studies: The ASSIST Approach

The current work addresses the unification of Electronic Health Records related to cervical cancer into a single medical knowledge source, in the context of the EU-funded ASSIST research project. The project aims to facilitate the research for cervical precancer and cancer through a system that virtually unifies multiple patient record repositories, physically located in different medical centers/hospitals, thus, increasing flexibility by allowing the formation of study groups “on demand” and by recycling patient records in new studies. To this end, ASSIST uses semantic technologies to translate all medical entities (such as patient examination results, history, habits, genetic profile) and represent them in a common form, encoded in the ASSIST Cervical Cancer Ontology. The current paper presents the knowledge elicitation approach followed, towards the definition and representation of the diseases medical concepts and rules that constitute the basis for the ASSIST Cervical Cancer Ontology. The proposed approach constitutes a paradigm for semantic integration of heterogeneous clinical data that may be applicable to other biomedical application domains.

On the use of higher-order statistics for robust endpoint detection of speech

Third order statistics of speech signals are not identically zero, as it would be expected based on the linear model for voice. This is due to quadratic harmonic coupling produced in the vocal tract. Based on this observation, third order cumulants are employed to address the endpoint detection problem in low SNR level recordings due to their immunity to (colored) additive non-skewed noise. The proposed method uses the maximum singular value of an appropriately formed cumulant matrix to distinguish between voiced parts of the speech signal, and silence (noise). Adaptive implementations are also proposed, making this method computationally attractive. Results of batch and adaptive forms are presented for real and simulated data.<<ETX>>

Efficient Quantitative Information Extraction from PCR-RFLP Gel Electrophoresis Images

For the purpose of PCR-RFLP analysis, as in the case of human papillomavirus (HPV) typing, quantitative information needs to be extracted from images resulting from one-dimensional gel electrophoresis by associating the image intensity with the concentration of biological material at the corresponding position on a gel matrix. However, the background intensity of the image stands in the way of quantifying this association. We propose a novel, efficient methodology for modeling the image background with a polynomial function and prove that this can benefit the extraction of accurate information from the lane intensity profile when modeled by a superposition of properly shaped parametric functions.

Least squares estimation of 3D shape and motion of rigid objects from their orthographic projections

The extraction of motion and shape information of three-dimensional objects from their two-dimensional projections is a task that emerges in various applications such as computer vision, biomedical engineering, and video coding and mining especially after the recent guidelines of the Motion Pictures Expert Group regarding MPEG-4 and MPEG-7 standards. Present work establishes a novel approach for extracting the motion and shape parameters of a rigid three-dimensional object on the basis of its orthographic projections and the associated motion field. Experimental results have been included to verify the theoretical analysis.