Ricardo José Ferrari

Automatic identification of the pectoral muscle in mammograms

The pectoral muscle represents a predominant density region in most medio-lateral oblique (MLO) views of mammograms; its inclusion can affect the results of intensity-based image processing methods or bias procedures in the detection of breast cancer. Local analysis of the pectoral muscle may be used to identify the presence of abnormal axillary lymph nodes, which may be the only manifestation of occult breast carcinoma. We propose a new method for the identification of the pectoral muscle in MLO mammograms based upon a multiresolution technique using Gabor wavelets. This new method overcomes the limitation of the straight-line representation considered in our initial investigation using the Hough transform. The method starts by convolving a group of Gabor filters, specially designed for enhancing the pectoral muscle edge, with the region of interest containing the pectoral muscle. After computing the magnitude and phase images using a vector-summation procedure, the magnitude value of each pixel is propagated in the direction of the phase. The resulting image is then used to detect the relevant edges. Finally, a post-processing stage is used to find the true pectoral muscle edge. The method was applied to 84 MLO mammograms from the Mini-MIAS (Mammographic Image Analysis Society, London, U.K.) database. Evaluation of the pectoral muscle edge detected in the mammograms was performed based upon the percentage of false-positive (FP) and false-negative (FN) pixels determined by comparison between the numbers of pixels enclosed in the regions delimited by the edges identified by a radiologist and by the proposed method. The average FP and FN rates were, respectively, 0.58% and 5.77%. Furthermore, the results of the Gabor-filter-based method indicated low Hausdorff distances with respect to the hand-drawn pectoral muscle edges, with the mean and standard deviation being 3.84/spl plusmn/1.73 mm over 84 images.

Analysis of asymmetry in mammograms via directional filtering with Gabor wavelets

This paper presents a procedure for the analysis of left-right (bilateral) asymmetry in mammograms. The procedure is based upon the detection of linear directional components by using a multiresolution representation based upon Gabor wavelets. A particular wavelet scheme with two-dimensional Gabor filters as elementary functions with varying tuning frequency and orientation, specifically designed in order to reduce the redundancy in the wavelet-based representation, is applied to the given image. The filter responses for different scales and orientation are analyzed by using the Karhunen-Loeve (KL) transform and Otsus method of thresholding. The KL transform is applied to select the principal components of the filter responses, preserving only the most relevant directional elements appearing at all scales. The selected principal components, thresholded by using Otsus method, are used to obtain the magnitude and phase of the directional components of the image. Rose diagrams computed from the phase images and statistical measures computed thereof are used for quantitative and qualitative analysis of the oriented patterns. A total of 80 images from 20 normal cases, 14 asymmetric cases, and six architectural distortion cases from the Mini-MIAS (Mammographic Image Analysis Society, London, U.K.) database were used to evaluate the scheme using the leave-one-out methodology. Average classification accuracy rates of up to 74.4% were achieved.

Identification of the breast boundary in mammograms using active contour models

A method for the identification of the breast boundary in mammograms is presented. The method can be used in the preprocessing stage of a system for computeraided diagnosis (CAD) of breast cancer and also in the reduction of image file size in picture archiving and communication system applications. The method started with modification of the contrast of the original image. A binarisation procedure was then applied to the image, and the chain-code algorithm was used to find an approximate breast contour. Finally, the identification of the true breast boundary was performed by using the approximate contour as the input to an active contour model algorithm specially tailored for this purpose. After demarcation of the breast boundary, all artifacts outside the breast region were eliminated. The method was applied to 84 medio-lateral oblique mammograms from the Mini-MIAS database. Evaluation of the detected breast boundary was performed based upon the percentage of false-positive and false-negative pixels determined by a quantitative comparison between the contours identified by a radiologist and those identified by the proposed method. The average false positive and false negative rates were 0.41% and 0.58%, respectively. The two radiologists who evaluated the results considered the segmentation results to be acceptable for CAD purposes.

Segmentation of the fibro-glandular disc in mammograms using Gaussian mixture modelling.

The paper presents a technique for the segmentation of the fibro-glandular disc in mammograms based upon a statistical model of breast density. The density function of the model was represented by a mixture of up to four weighted Gaussians, each one corresponding to a specific density class in the breast. The parameters of the model and the number of tissue classes in the breast were determined using the expectation-maximisation algorithm and the minimum description length method. Grey-level statistics of the pectoral muscle were used to determine the tissue categories that are likely to represent the fibro-glandular disc. The method was applied to 84 medio-lateral oblique mammograms from the Mini-MIAS database. The results of the segmented fibro-glandular disc were assessed by a radiologist using the original and the segmented images, with reference to a ranking table categorising the results of segmentation as: 1: excellent; 2: good; 3: average; 4: poor; and 5: complete failure. Of the 84 cases analysed, 64.3% were rated as excellent, 16.7% were rated as good, 10.7% were rated as average, and 4.7% were rated as poor; only 3.6% of the cases were rated as a complete failure with regard to segmentation of the fibro-glandular disc.

Real-time detection of steam in video images

In this paper, we present a real-time image processing technique for the detection of steam in video images. The assumption made is that the presence of steam acts as a blurring process, which changes the local texture pattern of an image while reducing the amount of details. The problem of detecting steam is treated as a supervised pattern recognition problem. A statistical hidden Markov tree (HMT) model derived from the coefficients of the dual-tree complex wavelet transform (DT-CWT) in small 48x48 local regions of the image frames is used to characterize the steam texture pattern. The parameters of the HMT model are used as an input feature vector to a support vector machine (SVM) technique, specially tailored for this purpose. By detecting and determining the total area covered by steam in a video frame, a computerized image processing system can automatically decide if the frame can be used for further analysis. The proposed method was quantitatively evaluated by using a labelled image data set with video frames sampled from a real oil sand video stream. The classification results were 90% correct when compared to human labelled image frames. The technique is useful as a pre-processing step in automated image processing systems.

Analysis of bilateral asymmetry in mammograms using directional, morphological, and density features

We propose techniques to analyze bilateral asymmetry in mammograms by combining directional information, morphological measures, and geometric moments related to density distributions. The procedure starts by detecting the breast boundary and the pectoral muscle edge (in mediolateral-oblique, or MLO, views). All artifacts outside the breast boundary as well as the pectoral muscle region are eliminated. A breast density model based upon a Gaussian mixture model is then used to segment the fibroglandular disks of the mammograms. Rose diagrams are used to map the magnitude and directional information related to the fibroglandular tissue filtered using multiresolution Gabor wavelets. The directional data of the left and right mammograms are aligned by using the straight lines perpendicular to the corresponding pectoral muscle edges and subtracted to obtain difference rose diagrams. Directional features are obtained from the difference rose diagrams and used to characterize the changes caused by the development of breast cancer in the form of bilateral asymmetry or architectural distortion. An additional set of features including Hus moments, eccentricity, stretch, area, and average density are extracted from the segmented fibroglandular disks. The differences between the pairs of the features for the left and right mammograms are used as measures for the analysis of asymmetry. The techniques were applied to 88 mammograms from the Mini-MIAS database. Classification accuracies of up to 84.4% were achieved, with sensitivity and specificity rates of 82.6% and 86.4%, respectively.

Directional analysis of images with Gabor wavelets

The paper presents a new scheme for analysis of linear directional components in images by using a multiresolution representation based on Gabor wavelets. A dictionary of Gabor filters with varying tuning frequency and orientation, specifically designed in order to reduce the redundancy in the wavelet-based representation, is applied to the given image. The filter responses for different scales and orientation are analyzed by using the Karhunen-Loeve (KL) transform and Otsus (1979) method of thresholding. The KL transform is applied to select the principal components of the filter responses, preserving only the most relevant directional elements appearing at all scales. The first N principal components, thresholded by using Otsus method, are used to reconstruct the magnitude and phase of the directional components of the image. Rose diagrams computed from the phase images are used for quantitative and qualitative analysis of the oriented patterns. The proposed scheme is applied to the analysis of asymmetry between left and right mammograms. For this purpose, a set of three features is extracted from the Rose diagrams and used in a parametric statistical classifier. A total of 80 images from 20 normal cases, 14 asymmetric cases, and 6 distortion cases from the Mini-MIAS database were used to evaluate the scheme using the leave-one-out methodology, resulting in an average diagnostic accuracy of 72.5%.

Digital radiographic image denoising via wavelet-based hidden Markov model estimation.

This paper presents a technique for denoising digital radiographic images based upon the wavelet-domain Hidden Markov tree (HMT) model. The method uses the Anscombe’s transformation to adjust the original image, corrupted by Poisson noise, to a Gaussian noise model. The image is then decomposed in different subbands of frequency and orientation responses using the dual-tree complex wavelet transform, and the HMT is used to model the marginal distribution of the wavelet coefficients. Two different correction functions were used to shrink the wavelet coefficients. Finally, the modified wavelet coefficients are transformed back into the original domain to get the denoised image. Fifteen radiographic images of extremities along with images of a hand, a line-pair, and contrast–detail phantoms were analyzed. Quantitative and qualitative assessment showed that the proposed algorithm outperforms the traditional Gaussian filter in terms of noise reduction, quality of details, and bone sharpness. In some images, the proposed algorithm introduced some undesirable artifacts near the edges.

Segmentation of Multiple Sclerosis Lesions Using Support Vector Machines

In this paper we present preliminary results to automatically segment multiple sclerosis (MS) lesions in multispectral magnetic resonance datasets using support vector machines (SVM). A total of eighteen studies (each composed of T1-, T2-weighted and FLAIR images) acquired from a 3T GE Signa scanner was analyzed. A neuroradiologist used a computer-assisted technique to identify all MS lesions in each study. These results were used later in the training and testing stages of the SVM classifier. A preprocessing stage including anisotropic diffusion filtering, non-uniformity intensity correction, and intensity tissue normalization was applied to the images. The SVM kernel used in this study was the radial basis function (RBF). The kernel parameter (γ) and the penalty value for the errors were determined by using a very loose stopping criterion for the SVM decomposition. Overall, a 5-fold cross-validation accuracy rate of 80% was achieved in the automatic classification of MS lesion voxels using the proposed SVM-RBF classifier.

Off-line determination of the optimal number of iterations of the robust anisotropic diffusion filter applied to denoising of brain MR images

Although anisotropic diffusion filters have been used extensively and with great success in medical image denoising, one limitation of this iterative approach, when used on fully automatic medical image processing schemes, is that the quality of the resulting denoised image is highly dependent on the number of iterations of the algorithm. Using many iterations may excessively blur the edges of the anatomical structures, while a few may not be enough to remove the undesirable noise. In this work, a mathematical model is proposed to automatically determine the number of iterations of the robust anisotropic diffusion filter applied to the problem of denoising three common human brain magnetic resonance (MR) images (T1-weighted, T2-weighted and proton density). The model is determined off-line by means of the maximization of the mean structural similarity index, which is used in this work as metric for quantitative assessment of the resulting processed images obtained after each iteration of the algorithm. After determining the model parameters, the optimal number of iterations of the algorithm is easily determined without requiring any extra computation time. The proposed method was tested on 3D synthetic and clinical human brain MR images and the results of qualitative and quantitative evaluation have shown its effectiveness.