Antonio Fernando Catelli Infantosi

Analysis of Co-Occurrence Texture Statistics as a Function of Gray-Level Quantization for Classifying Breast Ultrasound

In this paper, we investigated the behavior of 22 co-occurrence statistics combined to six gray-scale quantization levels to classify breast lesions on ultrasound (BUS) images. The database of 436 BUS images used in this investigation was formed by 217 carcinoma and 219 benign lesions images. The region delimited by a minimum bounding rectangle around the lesion was employed to calculate the gray-level co-occurrence matrix (GLCM). Next, 22 co-occurrence statistics were computed regarding six quantization levels (8, 16, 32, 64, 128, and 256), four orientations (0° , 45° , 90° , and 135° ), and ten distances (1, 2,...,10 pixels). Also, to reduce feature space dimensionality, texture descriptors of the same distance were averaged over all orientations, which is a common practice in the literature. Thereafter, the feature space was ranked using mutual information technique with minimal-redundancy-maximal-relevance (mRMR) criterion. Fisher linear discriminant analysis (FLDA) was applied to assess the discrimination power of texture features, by adding the first m-ranked features to the classification procedure iteratively until all of them were considered. The area under ROC curve (AUC) was used as figure of merit to measure the performance of the classifier. It was observed that averaging texture descriptors of a same distance impacts negatively the classification performance, since the best AUC of 0.81 was achieved with 32 gray levels and 109 features. On the other hand, regarding the single texture features (i.e., without averaging procedure), the quantization level does not impact the discrimination power, since AUC=0.87 was obtained for the six quantization levels. Moreover, the number of features was reduced (between 17 and 24 features). The texture descriptors that contributed notably to distinguish breast lesions were contrast and correlation computed from GLCMs with orientation of 90° and distance more than five pixels.

Complexity curve and grey level co‐occurrence matrix in the texture evaluation of breast tumor on ultrasound images

This work aims at investigating texture parameters in distinguishing malign and benign breast tumors on ultrasound images. A rectangular region of interest (ROI) containing the tumor and its neighboring was defined for each image. Five parameters were extracted from the complexity curve (CC) of the ROI. Another five parameters were calculated from the grey-level co-occurrence matrix (GLCM) also for the ROI. The same was carried out for internal tumor region, hence, totaling 20 parameters. The linear discriminant analysis was applied to sets of up to five parameters and then the performances were assessed. The most relevant individual parameters were the contrast (con) (from the GLCM over the ROI) and the maximum value (mvi) from the CC just for the tumor internal region). When they were taken together, a correct classification slightly over 80% of the breast tumors was achieved. The highest performance (accuracy=84.2%, sensitivity=87.0%, and specificity=78.8%) was obtained with mvi, con, the standard deviation of the pixel pairs and the entropy, both for GLCM, and the internal region contrast also from GLCM. Parameters extracted from the internal region generally performed better and were more significant than those from the ROI. Moreover, parameters calculated only from CC or GLCM resulted in no statistically significant performance difference. These findings suggest that the texture parameters can be useful to help radiologist in distinguishing between benign or malign breast tumors on ultrasound images.

Objective response detection in an electroencephalogram during somatosensory stimulation.

AbstractTechniques for objective response detection aim to identify the presence of evoked potentials based purely on statistical principles. They have been shown to be potentially more sensitive than the conventional approach of subjective evaluation by experienced clinicians and could be of great clinical use. Three such techniques to detect changes in an electroencephalogram (EEG) synchronous with the stimuli, namely, magnitude-squared coherence (MSC), the phase-synchrony measure (PSM) and the spectral F test (SFT) were applied to EEG signals of 12 normal subjects under conventional somatosensory pulse stimulation to the tibial nerve. The SFT, which uses only the power spectrum, showed the poorest performance, while the PSM, based only on the phase spectrum, gave results almost as good as those of the MSC, which uses both phase and power spectra. With the latter two techniques, stimulus responses were evident in the frequency range of 20–80 Hz in all subjects after 200 stimuli (5 Hz stimulus frequency), whereas for visual recognition at least 500 stimuli are usually applied. Based on these results and on simulations, the phase-based techniques appear promising for the automated detection and monitoring of somatosensory evoked potentials.

Coherence between one random and one periodic signal for measuring the strength of responses in the electro-encephalogram during sensory stimulation

Coherence between a pulse train representing periodic stimuli and the EEG has been used in the objective detection of steady-state evoked potentials. This work aimed to quantify the strength of the stimulus responses based on the statistics of coherence estimate between one random and one periodic signal, focusing on the confidence limits and power of significance tests in detecting responses. To detect the responses in 95% of cases, a signal-to-noise ratio of about −7.9 dB was required when using 48 windows (M) in the coherence estimation. The ratio, however, increased to −1.2 dB when M was 12. The results were tested in Monte Carlo simulations and applied to EEGs obtained from 14 subjects during visual stimulation. The method showed differences in the strength of responses at the stimulus frequency and its harmonics, as well as variations between individuals and over cortical regions. In contrast to those from the parietal and temporal regions, results for the occipital region gave confidence limits (with M=12) that were above zero for all subjects, indicating statistically significant responses. The proposed technique extends the usefulness of coherence as a measure of stimulus responses and allows statistical analysis that could also be applied usefully in a range of other biological signals.

Assessing the performance of morphological parameters in distinguishing breast tumors on ultrasound images

This work aims at investigating seven morphological parameters in distinguishing malignant and benign breast tumors on ultrasound images. Linear discriminant analysis was applied to sets of up to five parameters and then the performances were assessed using the area Az (+/- standard error) under the ROC curve, accuracy (Ac), sensitivity (Se), specificity (Sp), positive predictive value and negative predictive value. The most relevant individual parameters were the normalized residual value (nrv) and overlap ratio (RS), both calculated from the convex polygon technique, and the circularity (C). When nrv and C were taken together with roughness (R), calculated from normalized radial length (NRL), a performance slightly over 83% in distinguishing malignant and benign breast tumors was achieved.

Classification of breast tumours on ultrasound images using morphometric parameters

This work aims to assess the potentiality of morphometric parameters in separating breast tumour, on ultrasonic images, as malign or benign. Parameters were calculated over normalised radial length and convex polygons from 152 segmented tumour images. Linear discriminant analysis was applied and parameters performance assessed (accuracy, sensitivity and specificity). The best parameter performances for individual parameters were the normalised residual mean square value and the circularity. Taking these last two and the roughness the best separation performance was obtained: specificity (90.4%) and sensitivity (88.0%). These three parameters were also applied to a multilayer perceptron network using GA-backpropagation hybrid training. The initial results pointed out that hybrid GA-backpropagation training was capable to produce similar high performance both to training (accuracy = 90.3%, sensitivity = 90.0% and specificity = 90.9%) and test (accuracy, sensitivity and specificity equal to 90.0%) procedures. Besides, the performances obtained with two training sets of distinct sizes (30% and 50% of all samples) were slightly different.

Estimation and significance testing of cross-correlation between cerebral blood flow velocity and background electro-encephalograph activity in signals with missing samples

Cross-correlation between cerebral blood flow (CBF) and background EEG activity can indicate the integrity of CBF control under changing metabolic demand. The difficulty of obtaining long, continuous recordings of good quality for both EEG and CBF signals in a clinical setting is overcome, in the present work, by an algorithm that allows the cross-correlation function (CCF) to be estimated when the signals are interrupted by segments of missing data. Methods are also presented to test the statistical significance of the CCF obtained in this way and to estimate the power of this test, both based on Monte Carlo simulations. The techniques are applied to the time-series given by the mean CBF velocity (recorded by transcranial Doppler) and the mean power of the EEG signal, obtained in 1 s intervals from nine sleeping neonates. The peak of the CCF is found to be low (≤0.35), but reached statistical significance (p<0.05) in five of the nine subjects. The CCF further indicates a delay of 4–6s between changes in EEG and CBF velocity. The proposed signal-analysis methods prove effective and convenient and can be of wide use in dealing with the common problem of missing samples in biological signals.

Improving classification performance of breast lesions on ultrasonography

Several morphological and texture features aiming to distinguish between benign and malignant lesions on breast ultrasound (BUS) have been proposed in the literature. Various authors also claim that their particular feature sets are capable of reaching adequate classification rate. However, there are still several features that have not been tested together for determining the feature set that effectively improves classification performance. Hence, in this paper, we compiled distinct morphological and texture features widely used in computer-aided diagnosis systems for BUS images. A total of 26 morphological and 1465 texture features were computed from 641 BUS images (413 benign and 228 malignant lesions). A feature selection methodology, based on mutual information and statistical tests, was used to evaluate the discrimination power of distinct feature subsets. The .632+ bootstrap method was used to estimate the classification performance of each feature subset, by using the local Fisher discriminant analysis (LFDA), with linear kernel, as classifier, and the area under ROC curve (AUC) as performance index. The experimental results indicated that the best classification performance is AUC=0.942, obtained by a morphological set with five features. In addition, this morphological set outperformed the best texture set with four features, which attained AUC=0.897. The classification performances of 11 feature sets proposed in the literature were also surpassed by such morphological feature set. Highlights1491 features are evaluated for classifying breast lesions on ultrasound.Feature selection is based on mutual information and statistical tests.5 morphological and 4 texture features achieve the best classification performance.11 feature sets from the literature are surpassed by the 5 morphological features.

Modeling extracellular space electrodiffusion during Lea/spl tilde/o's spreading depression

Computational modeling of spreading depression (SD) has been used increasingly to study the different mechanisms that are involved in this phenomenon. One of them that is still under discussion involves the mechanisms that originate the extracellular electrical field responsible for the dc potential shift. The main goal of this paper is to present a mathematical derivation for the extracellular electric field that is incorporated in a SD model that has the basic structure of Tuckwell and Miuras model, but with the ionic variations calculated electrochemically. Electrodiffusion equations were used to describe the ionic movement of the four ions Na/sup +/, K/sup +/, Cl/sup -/, and Ca/sup 2+/. These are mutually coupled by the electric field within the extracellular space (ECS). The results from the simulations show that the model is able to calculate the effect of the ionic changes along the ECS on the electric field, and to reproduce the SD in respect to the most important features that characterize the phenomenon experimentally in the retina or hippocampus. It is suggested that the extracellular negative field-potential shift during SD is due to an electrical field generated by a Goldman-Hodgkin-Katz equation acting within the ECS.

Model and simulation of Na+/K+ pump phosphorylation in the presence of palytoxin

The ATP hydrolysis reactions responsible for the Na(+)/K(+)-ATPase phosphorylation, according to recent experimental evidences, also occur for the PTX-Na(+)/K(+) pump complex. Moreover, it has been demonstrated that PTX interferes with the enzymes phosphorylation status. However, the reactions involved in the PTX-Na(+)/K(+) pump complex phosphorylation are not very well established yet. This work aims at proposing a reaction model for PTX-Na(+)/K(+) pump complex, with similar structure to the Albers-Post model, to contribute to elucidate the PTX effect over Na(+)/K(+)-ATPase phosphorylation and dephosphorylation. Computational simulations with the proposed model support several hypotheses and also suggest: (i) phosphorylation promotes an increase of the open probability of induced channels; (ii) PTX reduces the Na(+)/K(+) pump phosphorylation rate; (iii) PTX may cause conformational changes to substates where the Na(+)/K(+)-ATPase may not be phosphorylated; (iv) PTX can bind to substates of the two principal states E1 and E2, with highest affinity to phosphorylated enzymes and with ATP bound to its low-affinity sites. The proposed model also allows previewing the behavior of the PTX-pump complex substates for different levels of intracellular ATP concentrations.