Marcello Salmeri

Mammographic Images Enhancement and Denoising for Breast Cancer Detection Using Dyadic Wavelet Processing

Mammography is the most effective method for the early detection of breast diseases. However, the typical diagnostic signs such as microcalcifications and masses are difficult to detect because mammograms are low-contrast and noisy images. In this paper, a novel algorithm for image denoising and enhancement based on dyadic wavelet processing is proposed. The denoising phase is based on a local iterative noise variance estimation. Moreover, in the case of microcalcifications, we propose an adaptive tuning of enhancement degree at different wavelet scales, whereas in the case of mass detection, we developed a new segmentation method combining dyadic wavelet information with mathematical morphology. The innovative approach consists of using the same algorithmic core for processing images to detect both microcalcifications and masses. The proposed algorithm has been tested on a large number of clinical images, comparing the results with those obtained by several other algorithms proposed in the literature through both analytical indexes and the opinions of radiologists. Through preliminary tests, the method seems to meaningfully improve the diagnosis in the early breast cancer detection with respect to other approaches.

Metrological Characterization of a CADx System for the Classification of Breast Masses in Mammograms

In this paper, we perform the assessment of a CAD for the tumoral masses classification in mammograms by the uncertainty propagation through the system. Carrying on the work of the authors concerning the metrological characterization of the developed CAD, we validate the features extraction, features selection, and classification steps in this paper. In particular, suitable metrics such as the Receiving Operating Curve (ROC) and the Area Under ROC (AUC) are widely used in order to provide a quantitative evaluation of the performance. Finally, we implement a Monte Carlo simulation in order to provide the confidence interval for some coverage probabilities for all involved parameters. The procedure is tested on mammographic images containing both malignant and benign breast masses.

Noise estimation in digital images using fuzzy processing

Noise estimation is an important issue in image processing because it is a fundamental step in many algorithms for noise suppression and then for image restoration. In the literature, many approaches have been presented in order to obtain good results. This paper presents a novel method suitable for obtaining a good estimation if the type of noise distribution is known. In particular, the algorithm provides the variance of the noise distribution and the proof that the distribution itself matches the foreseen one. The algorithm has been tested on different images affected by Gaussian noise. The simulations show results better than those obtained with other approaches.

Analysis of Structural Similarity in Mammograms for Detection of Bilateral Asymmetry

We hypothesize that quantification of structural similarity or dissimilarity between paired mammographic regions can be effective in detecting asymmetric signs of breast cancer. Bilateral masking procedures are applied for this purpose by using automatically detected anatomical landmarks. Changes in structural information of the extracted regions are investigated using spherical semivariogram descriptors and correlation-based structural similarity indices in the spatial and complex wavelet domains. The spatial distribution of grayscale values as well as of the magnitude and phase responses of multidirectional Gabor filters are used to represent the structure of mammographic density and of the directional components of breast tissue patterns, respectively. A total of 188 mammograms from the DDSM and mini-MIAS databases, consisting of 47 asymmetric cases and 47 normal cases, were analyzed. For the combined dataset of mammograms, areas under the receiver operating characteristic curves of 0.83, 0.77, and 0.87 were obtained, respectively, with linear discriminant analysis, the Bayesian classifier, and an artificial neural network with radial basis functions, using the features selected by stepwise logistic regression and leave-one-patient-out cross-validation. Two-view analysis provided accuracy up to 0.94, with sensitivity and specificity of 1 and 0.88, respectively.

Estimation of the breast skin-line in mammograms using multidirectional Gabor filters

Segmentation of the breast region is a fundamental step in any system for computerized analysis of mammograms. In this work, we propose a novel procedure for the estimation of the breast skin-line based upon multidirectional Gabor filtering. The method includes an adaptive values-of-interest (VOI) transformation, extraction of the skin-air ribbon by Otsus thresholding method and the Euclidean distance transform, Gabor filtering with 18 real kernels, and a step for suppression of false edge points using the magnitude and phase responses of the filters. On a test set of 361 images from different acquisition modalities (screen-film and full-field digital mammograms), the average Hausdorff and polyline distances obtained were 2.85 mm and 0.84 mm, respectively, with reference to the ground-truth boundaries provided by an expert radiologist. When compared with the results obtained by other state-of-the-art methods on the same set of images and with respect to the same ground-truth boundaries, our method mostly outperformed the other approaches. The results demonstrate the effectiveness and robustness of the proposed algorithm.

Type-2 Fuzzy Sets for Modeling Uncertainty in Measurement

A correct representation of uncertainty in measurement is crucial in many applications. Statistical approach sometimes is not the best choice, especially when the knowledge of the measurement process refers only to the support of the values and does not allow a correct assumption on the probability density function (pdf) of the measured variable. In this paper we present an approach that uses the concept of generalized fuzzy numbers, namely type-2 fuzzy sets, in order to handle the intrinsic dispersion of the possible pdfs associated to a variable. The relation between our representation and the so called random fuzzy variables (RFV) will be also investigated. The use of this representation allows to easily implement the uncertainty propagation, through a functional model, by working directly on the type-2 fuzzy numbers and by evaluating simultaneously the propagation results for the whole set of confidence levels. Anyway, when a statistical analysis can be performed, the results can be embedded in this generalized representation. Moreover, the new approach allows to assign to the final measurement value a reliable confidence level also in this case, by combining the expanded uncertainty evaluated following IEC-ISO guide recommendations with the type-2 fuzzy numbers associated to the output variable. An example of this representation was provided

Breast masses detection using phase portrait analysis and fuzzy inference systems.

PurposeBreast masses exhibit variability in margins, shapes, and dimensions, so their detection is a difficult task in mammographic computer-aided diagnosis. Mass detection is usually a two-step procedure: mass identification and false-positive reduction. A new method to automatically detect mass lesions in mammographic images with tuning according to the breast tissue density was developed and tested.MethodsA modified phase portrait analysis method was introduced, based on the eigenvalue condition number and an eigenvalue intensity map. The method uses an iterative and tissue density-adaptive segmentation procedure with extraction of geometric features. False-positive reduction is accomplished using a fuzzy inference-based classifier. A leave-one-image-out cross-validation procedure was implemented, and stepwise regression analysis was used to automatically extract an optimal set of features. Testing and validation were performed on two different data sets containing at least one malignant mass D1 (388 images) and D2 (674 images), and a third data set N1 (50 images) was used consisting of normal controls. These three data sets were taken from the Digital Database for Screening Mammography.ResultsFor sensitivities of 0.9, 0.85, 0.80, and 0.75, the best results on cancer images exhibit an False-Positive per Image (FPpI) equal to 0.6, 0.45, 0.35, and 0.3, respectively, using a Bayes Linear Discriminant Analysis (LDA) classifier and an FPpI of 0.85, 0.7, 0.55, and 0.45 using a fuzzy inference system (FIS) for false-positive reduction. When the algorithm is tested on normal images, an FPpI equal to 0.4, 0.3, 0.25, and 0.2 was observed using LDA and 0.3, 0.25, 0.2, and 0.15 using the FIS.ConclusionA preclinical study of an automatic breast mass detection algorithm provided promising results in terms of sensitivity and low false-positive rate. Further development and clinical testing are justified based on the results.

Parallel hardware implementation of radar electronics equipment for a laser inspection system

An amplitude modulated laser radar has been developed by ENEA (Italian Agency for new technologies, energy and environment) for periodic in-vessel inspection in large fusion machines (ITER). The system is able to obtain a complete 3D mapping of the in-vessel surface. First, a digital signal processing system was developed to modulate the laser beam and to detect both the amplitude of the back scattered light and the phase difference between it and the modulation signal. This system is based on commercial digital receiver and parallel DSP (digital signal processing) boards on a VME bus. It reaches a speed of 100 K measures/s showing good accuracy and stability. Starting from this, further development has been done to increase the speed up to 2.328 M measures/s. To reach the sub-microsecond speed it was necessary to implement the mathematical algorithm in a highly parallel hardware architecture using FPGAs (field programmable gate array). Looking at the good results of previously developed system it was decided to maintain the same acquisition front-end. The last release of A/D converters was used to increase the operating frequency up to 200 MHz, but the previously used software algorithm was completely redesigned and optimized to be used in the FPGA hardware architecture.

Performance evaluation of a region growing procedure for mammographic breast lesion identification

At present, mammography is the most effective examination for an early diagnosis of breast cancer. Nevertheless, the detection of cancer signs in mammograms is a difficult procedure owing to the great number of non-pathological structures which are also present in the image. Recent statistics show that in current breast cancer screenings 10%-25% of the tumors are missed by the radiologists. For this reason, a lot of research is currently being done to develop systems for Computer Aided Detection (CADe). Probably, some causes of the false-negative screening examinations are that tumoral masses have varying dimension and irregular shape, their borders are often ill-defined and their contrast is very low, thus making difficult the discrimination from parenchymal structures. Therefore, in a CADe system a preliminary segmentation procedure has to be implemented in order to separate the mass from the background tissue. In this way, various characteristics of the segmented mass can be evaluated and used in a classification step to discriminate benign and malignant cases. In this paper, we describe an effective algorithm for massive lesions segmentation based on a region-growing technique and we provide full details the performance evaluation procedure used in this specific context.

DENOISING AND ENHANCEMENT OF MAMMOGRAPHIC IMAGES UNDER THE ASSUMPTION OF HETEROSCEDASTIC ADDITIVE NOISE BY AN OPTIMAL SUBBAND THRESHOLDING

Mammographic images suffer from low contrast and signal dependent noise, and a very small size of tumoral signs is not easily detected, especially for an early diagnosis of breast cancer. In this context, many methods proposed in literature fail for lack of generality. In particular, too weak assumptions on the noise model, e.g., stationary normal additive noise, and an inaccurate choice of the wavelet family that is applied, can lead to an information loss, noise emphasizing, unacceptable enhancement results, or in turn an unwanted distortion of the original image aspect. In this paper, we consider an optimal wavelet thresholding, in the context of Discrete Dyadic Wavelet Transforms, by directly relating all the parameters involved in both denoising and contrast enhancement to signal dependent noise variance (estimated by a robust algorithm) and to the size of cancer signs. Moreover, by performing a reconstruction from a zero-approximation in conjunction with a Gaussian smoothing filter, we are able to extract the background and the foreground of the image separately, as to compute suitable contrast improvement indexes. The whole procedure will be tested on high resolution X-ray mammographic images and compared with other techniques. Anyway, the visual assessment of the results by an expert radiologist will be also considered as a subjective evaluation.