Search

Gait rhythm disturbances due to abnormal strides indicate the degenerative mobility regulation of motor neurons affected by Parkinson's disease (PD). The aim of this work is to compute the approximate entropy (ApEn), normalized symbolic entropy (NSE), and signal turns count (STC) parameters for the measurements of stride fluctuations in PD. Generalized linear regression analysis (GLRA) and support vector machine (SVM) techniques were employed to implement nonlinear gait pattern classifications. The classification performance was evaluated in terms of overall accuracy, sensitivity, specificity, precision, Matthews correlation coefficient (MCC), and area under the receiver operating characteristic (ROC) curve. Our experimental results indicated that the ApEn, NSE, and STC parameters computed from the stride series of PD patients were all significantly larger (Wilcoxon rank-sum test: p

This paper presents a new method consisting of two stages for automatic detection and segmentation of coronary arteries in X-ray angiograms. In the first stage, multiscale Gabor filters are used to detect vessel structures in the angiograms. The results of multiscale Gabor filtering are compared with those obtained by applying multiscale methods based on the top-hat operator, Hessian matrix, and Gaussian matched filters. The performance of the vessel-detection methods is evaluated through the area ( A z ) under the receiver operating characteristic (ROC) curve. In the second stage, coronary arteries are segmented by binarizing the magnitude response of Gabor filters using a new thresholding method based on multiobjective optimization, which is compared with seven thresholding methods. Measures of sensitivity, specificity, accuracy, and positive predictive value are used to analyze the segmentation methods, by comparing the results to the ground-truth markings of the vessels drawn by a specialist. Finally, the proposed method is compared with seven state-of-the-art vessel segmentation methods. The result of vessel detection using multiscale Gabor filters demonstrated high accuracy with A z = 0.961 with a training set of 40 angiograms and A z = 0.952 with an independent test set of 40 angiograms. The results of vessel segmentation with the multiobjective thresholding method provided an average accuracy of 0.881 with the test set of angiograms.

Changes in the characteristics of retinal vessels such as width and tortuosity can be signs of the presence of several diseases such retinopathy of prematurity (ROP) and diabetic retinopathy. Plus disease is an indicator of ROP which requires treatment and is signified by an increase in posterior venular width. In this work, we present image processing techniques for the detection, segmentation, tracking, and measurement of the width of the major temporal arcade (MTA), which is the thickest venular branch in the retina. Several image processing techniques have been employed, including the use of Gabor filters to detect the MTA, morphological image processing to obtain its skeleton, Canny's method to detect and select MTA vessel-edge candidates, least-squares fitting to interpolate the MTA edges, and geometrical procedures to measure the width of the MTA. The results, obtained using 110 retinal fundus images of preterm infants, indicate a statistically highly significant difference in MTA width of normal cases as compared to cases with plus disease ( p < 0.01 ) . The results provide good accuracy in computer-aided diagnosis (CAD) of plus disease with an area under the receiver operating characteristic curve of 0.76. The proposed methods may be used in CAD of plus disease and timely treatment of ROP in a clinical or teleophthalmological setting. HighlightsWe present techniques for measurement of the width of the major temporal arcade.Gabor filters and morphological image processing methods are applied.The methods can detect plus disease and retinopathy of prematurity.Area under the receiver operating characteristic curve of 0.76 obtained.

Detection of the nipple in mammograms is an important step in algorithms for the detection of breast cancer. However, locating the nipple position is a challenging task due to distortion and displacement of the nipple by breast diseases, improper imaging techniques, and variation of the characteristics of breast tissues with different imaging protocols or modalities. This paper presents a novel approach for the detection of the nipple in mammograms based on the converging characteristics of oriented patterns of the breast tissues towards the nipple. The oriented structures are extracted with a bank of real Gabor filters and are transformed into the Radon domain to analyze linear structures of tissue patterns to determine the nipple position. The performance of the method was evaluated with different types of images, such as scanned screen-film (from the mini-MIAS and DDSM databases), digital radiography, and computed radiography, and average errors of 7.71 mm, 7.52 mm, 9.23 mm, and 12.10 mm were achieved, respectively, with reference to the nipple location marked by an expert radiologist. The proposed method outperforms two recently developed approaches for the same application.

Concerns about the specificity and reliability of artificial neural networks (ANNs) impede further application of ANNs in medicine. This is particularly true when developing computer-aided diagnosis (CAD) tools using ANNs for orphan diseases and emerging research areas where only a small-sized sample set is available. It is unreasonable to claim one ANN's performance as better than another simply on the basis of a single output without considering possible output variability due to factors including data noise and ANN training protocols. In this paper, a bootstrap resampling method is proposed to quantitatively analyze ANN output reliability and changing performance as the sample data and training protocols are varied. The method is tested in the area of feature classification for analysis of masses detected on mammograms. Our experiments show that ANNs performance, measured in terms of the area under the receiver operating characteristic (ROC) curve, is not a fixed value, but follows a distribution function sensitive to many factors. We demonstrate that our approach to determining the bootstrap estimates of confidence intervals (CIs) and prediction intervals (PIs) can be used to assure optimal performance in terms of ANN model configuration. We also show that the unintentional inclusion of data noise, which biases ANN results in small task-specific databases, can be accurately detected via the bootstrap estimates.

Parameters useful for the diagnosis of pathological processes leading to the deterioration of the articular cartilage surfaces of knee joints, such as osteoarthritis, may be derived from vibroarthrographic (VAG) signals. In the present work, we explore fractal analysis to parameterize the temporal and spectral variability of normal and abnormal VAG signals. The power spectrum analysis method was used with the 1/ f model to derive estimates of the fractal dimension (FD). Classification accuracy of up to 0.74 was obtained with a single FD parameter, in terms of the area under the receiver operating characteristic curve ( A z ), with a database of 89 VAG signals. Combinations of the features derived in the present work with other features we have reported upon recently, when used with several neural networks with radial basis functions, resulted in A z values in the range [0.92, 0.96], with an exceptional case of perfect classification with A z = 1.0. The proposed methods could help in the detection and monitoring of knee-joint pathology.

Pathological conditions of knee joints have been observed to cause changes in the characteristics of vibroarthrographic (VAG) signals. Several studies have proposed many parameters for the analysis and classification of VAG signals; however, no statistical modeling methods have been explored to analyze the distinctions in the probability density functions (PDFs) between normal and abnormal VAG signals. In the present work, models of PDFs were derived using the Parzen-window approach to represent the statistical characteristics of normal and abnormal VAG signals. The Kullback-Leibler distance was computed between the PDF of the signal to be classified and the PDF models for normal and abnormal VAG signals. Additional statistical measures, including the mean, standard deviation, coefficient of variation, skewness, kurtosis, and entropy, were also derived from the PDFs obtained. An overall classification accuracy of 77.53%, sensitivity of 71.05%, and specificity of 82.35% were obtained with a database of 89 VAG signals using a neural network with radial basis functions with the leave-one-out procedure for cross validation. The screening efficiency was derived to be 0.8322, in terms of the area under the receiver operating characteristics curve.

Data-mining methods can be used to generate rules, or identify patterns, from medical data to assist clinical diagnosis and decision-making. However, in the initial stages of a clinical study on a new diagnostic approach, there could be a limited medical dataset available; or the medical characteristics could mean that the number of patients involved in the study will never be large. Diagnoses made using the rules discovered from such small medical databases should be considered suspect unless a confidence range for a particular diagnosis can be established. A method to evaluate the sensitivity and reliability of data-mining with small databases is presented in this paper. Efron's bootstrap method for statistical testing was used to assess the accuracy of the rules produced during the training step of the data-mining algorithm. The case study for validating this new approach was based on a limited-sized mammographic database previously used to discover associations between the diagnostic features of breast masses in mammograms and the biopsy-based classification of the masses. Using the new approach, it was possible to distinguish between the association rules that were sensitive to the size of the training datasets from those that were not. The methods proposed should lead to an efficient way for validating the patterns discovered in medical data-mining applications using small datasets.

We propose the strict 2-surface proximal (S2SP) classifier, by seeking two cross proximal planes to fit the distribution of the given samples in a corresponding feature space. The method is applied to screen knee-joint vibration or vibroarthrographic (VAG) signals based on statistical parameters derived from signals and selected by the genetic algorithm. A database of 89 VAG signals was studied. With the leave-one-out procedure, the linear S2SP classifier provided an efficiency of 0.82 in terms of the area under the receiver operating characteristics curve (A(z)); the nonlinear S2SP classifier provided 0.95 in A(z) value using the Gaussian kernel, and possessed good robustness around the selected kernel parameter.

Mammography is the best available tool for screening for the early detection of breast cancer. Mammographic screening has been shown to be effective in reducing breast cancer mortality rates: screening programs have reduced mortality rates by 30–70%. Mammograms are difficult to interpret, especially in the screening context. The sensitivity of screening mammography is affected by image quality and the radiologist's level of expertise. Computer-aided diagnosis (CAD) technology can improve the performance of radiologists, by increasing sensitivity to rates comparable to those obtained by double reading, in a cost-effective manner. Current research is directed toward the development of digital imaging and image analysis systems that can detect mammographic features, classify them, and provide visual prompts to the radiologist. Radiologists would like the ability to change the contrast of a mammogram, either manually or with pre-selected settings. Computer techniques for detecting, classifying, and annotating diagnostic features on the images would be desirable. This paper presents an overview of digital image processing and pattern analysis techniques to address several areas in CAD of breast cancer, including: contrast enhancement, detection and analysis of calcifications, detection and analysis of masses and tumors, analysis of bilateral asymmetry, and detection of architectural distortion. Although a few commercial CAD systems have been released, the detection of subtle signs of breast cancer such as global bilateral asymmetry and focal architectural distortion remains a difficult problem. We present some of our recent works on the development of image processing and pattern analysis techniques for these applications.

Segmentation of the tumor in computed tomographic (CT) images of neuroblastoma is a difficult problem due to its heterogeneity. In this paper, we present a survey of the nature, diagnosis, and management of the disease, and explore the potential application of image processing methods for the segmentation and analysis of the tumor. In particular, we study the use of opening-by-reconstruction and fuzzy connectivity to guide the segmentation process. We also investigate various strategies to improve the resulting tumor definition, including prior delineation of peripheral artifacts, peripheral fat, peripheral muscle, and the spinal canal, as well as postprocessing by morphological methods and deformable contours. Results of application of four procedures to 10 CT exams of four patients are analyzed. It is shown that opening-by-reconstruction is faster (by a factor of about 19) and provides a higher true-positive rate (83% versus 78%) than fuzzy connectivity. Leakage of the result of segmentation into neighboring structures remains to be a problem in cases where the tumor is significantly heterogeneous. Regardless, image processing methods could play an important role in improved management of neuroblastoma.

We propose methods for the analysis and classification of breast masses in mammograms via fractal analysis of signatures of their contours. Fractal dimension, on its own or in conjunction with other shape factors such as fractional concavity and spiculation index, resulted in the classification of mammographic masses as benign or malignant with overall accuracies of up to 0.93 with a database of 111 contours, as indicated by the area under the receiver operating characteristics curve. Furthermore, fractal dimension was able to classify atypical spiculated benign masses more accurately than other shape factors.

We propose a novel approach to detect frequency-modulated (FM) components such as chirp, sinusoidal FM, and hyperbolic FM components in multicomponent, nonstationary signals by application of pattern recognition techniques in the time–frequency (TF) plane. The approach, based upon the use of a combined Hough and Radon transform (HRT) can be used to detect FM components with varying energy and duration. The HRT method uses an array of accumulators, and the quantization of the array bins makes the method flexible and computationally attractive. Results with synthetic signals and a real-world signal (knee-joint vibration signal) indicate the potential of the method in identifying the components in noisy environments. The HRT method could be extended to detect components generated by any FM law provided they can be modeled by parametric equations.

Homomorphic filtering using the complex cepstrum has been used in many applications of one-dimensional (1D) signal processing, notably speech and seismic signal analysis. Some work has been done in extending cepstral filtering to the 2D domain. An application that we are particularly interested in is 2D echo removal. We show in this paper that 1D cepstral filtering techniques may be applied to the projections of images in the Radon domain. The filtered projections may then be used to reconstruct an image without the echoes. Results with synthetic images are presented to illustrate the strengths of the method. The technique could find applications in analysis of textured images and images obtained via reconstruction from projections.

This paper proposes non-invasive techniques to localize sound or vibroarthrographic (VAG) signal sources in human knee joints. VAG signals from normal subjects, patients who subsequently underwent arthroscopy, and cadavers with arthroscopically-created lesions, obtained by stimulation with a finger tap over the mid-patella and swinging movement of the leg, were analyzed for time delays using cross-correlation functions for source localization. Correct results were obtained for 13 of the 14 subjects tested by finger stimulation, and for 11 of the 12 subjects whose VAG signals during swinging movement were analyzed. The techniques could be valuable in the diagnosis and treatment of knee pathology before and after joint surgery or drug therapy.

By modifying the histogram of an image, a dramatic improvement in the perceptibility of details can often be achieved. However, the two commonly used methods of full-frame histogram equalization and local-area histogram equalization often fail to produce adequate enhancement when the image contains relatively small but variable-sized regions in which there are objects or features of interest with low visual contrast. A new method of adaptive-neighborhood histogram equalization that is effective in enhancing these types of images is proposed in this paper. In this method, an adaptive neighborhood is developed for each pixel in the image. The adaptive neighborhood is a compound region made up of a foreground that contains 8-connected pixels close in gray level to that of the seed pixel, and a background of neighboring pixels molded around the foreground. The histogram of this adaptive neighborhood is equalized to provide the transformation that is applied to the seed pixel. Major advantages of this method are the avoidance of block edge artifacts that are encountered in local-area histogram equalization, and improved perceptibility of image detail. Examples of images transformed using the three methods of histogram modification are presented along with a discussion of the merits of the adaptive-neighborhood method.

Methods are developed for producing 3-dimensional displays of anisotropic data arising from a series of limited-view computed tomograms. Volume data are formed by stacking 2-dimensional images reconstructed from projections from as few as three angles. We specify a slab of the volume from which parallel-ray projection images are produced. Two reprojection images are made in directions separated by a small angle. These form a stereo pair and when viewed together give an illusion of depth (3D display). By taking viewing angles near the angles used for the initial (data) projections, a few sharp, 3-dimensional views into the volume are obtained despite the anisotropy of the reconstruction. Also it is shown how image quality varies with the angle of view and thickness of the region.