Nikos Dimitropoulos

A fully automated scheme for mammographic segmentation and classification based on breast density and asymmetry

This paper presents a fully automated segmentation and classification scheme for mammograms, based on breast density estimation and detection of asymmetry. First, image preprocessing and segmentation techniques are applied, including a breast boundary extraction algorithm and an improved version of a pectoral muscle segmentation scheme. Features for breast density categorization are extracted, including a new fractal dimension-related feature, and support vector machines (SVMs) are employed for classification, achieving accuracy of up to 85.7%. Most of these properties are used to extract a new set of statistical features for each breast; the differences among these feature values from the two images of each pair of mammograms are used to detect breast asymmetry, using an one-class SVM classifier, which resulted in a success rate of 84.47%. This composite methodology has been applied to the miniMIAS database, consisting of 322 (MLO) mammograms -including 15 asymmetric pairs of images-, obtained via a (noisy) digitization procedure. The results were evaluated by expert radiologists and are very promising, showing equal or higher success rates compared to other related works, despite the fact that some of them used only selected portions of this specific mammographic database. In contrast, our methodology is applied to the complete miniMIAS database and it exhibits the reliability that is normally required for clinical use in CAD systems.

Multi-scaled morphological features for the characterization of mammographic masses using statistical classification schemes

OBJECTIVE A comprehensive signal analysis approach on the mammographic mass boundary morphology is presented in this article. The purpose of this study is to identify efficient sets of simple yet effective shape features, employed in the original and multi-scaled spectral representations of the boundary, for the characterization of the mammographic mass. These new methods of mass boundary representation and processing in more than one domain greatly improve the information content of the base data that is used for pattern classification purposes, introducing comprehensive spectral and multi-scale wavelet versions of the original boundary signals. The evaluation is conducted against morphological and diagnostic characterization of the mass, using statistical methods, fractal dimension analysis and a wide range of classifier architectures. METHODS AND MATERIALS This study consists of (a) the investigation of the original radial distance measurements under the complete spectrum of signal analysis, (b) the application of curve feature extractors of morphological characteristics and the evaluation of the discriminative power of each one of them, by means of statistical significance analysis and dataset fractal dimension, and (c) the application of a wide range of classifier architectures on these morphological datasets, in order to conduct a comparative evaluation of the efficiency and effectiveness of all architectures, for mammographic mass characterization. Radial distance signal was exploited using the discrete Fourier transform (DFT) and the discrete wavelet transform (DWT) as additional carrier signals. Seven uniresolution feature functions were applied over these carrier signals and multiple shape descriptors were created. Classification was conducted against mass shape type and clinical diagnosis, using a wide range of linear and non-linear classifiers, including linear discriminant analysis (LDA), least-squares minimum distance (LSMD), k-nearest neighbor (k-NN), radial basis function (RBF) and multi-layered perceptron (MLP) neural networks (NN), and support vector machines (SVM). Fractal analysis was employed as a dataset analysis tool in the feature selection phase. The discriminative power of the features produced by this composite analysis is subsequently analyzed by means of multivariate analysis of variance (MANOVA) and tested against two distinct classification targets, namely (a) the morphological shape type of the mass and (b) the histologically verified clinical diagnosis for each mammogram. RESULTS Statistical analysis and classification results have shown that the discrimination value of the features extracted from the DWT components and especially the DFT spectrum, are of great importance. Furthermore, much of the information content of the curve features in the case of DFT and DWT datasets is directly related to the texture and fine-scale details of the corresponding envelope signal of the spectral components. Neural classifiers outperformed all other methods (SVM not used because they are mainly two-class classifiers) with overall success rate of 72.3% for shape type identification, while SVM achieved the overall highest 91.54% for clinical diagnosis. Receiver operating characteristic (ROC) analysis has been employed to present the sensitivity and specificity of the results of this study.

Morphological and wavelet features towards sonographic thyroid nodules evaluation

This paper presents a computer-based classification scheme that utilized various morphological and novel wavelet-based features towards malignancy risk evaluation of thyroid nodules in ultrasonography. The study comprised 85 ultrasound images-patients that were cytological confirmed (54 low-risk and 31 high-risk). A set of 20 features (12 based on nodules boundary shape and 8 based on wavelet local maxima located within each nodule) has been generated. Two powerful pattern recognition algorithms (support vector machines and probabilistic neural networks) have been designed and developed in order to quantify the power of differentiation of the introduced features. A comparative study has also been held, in order to estimate the impact speckle had onto the classification procedure. The diagnostic sensitivity and specificity of both classifiers was made by means of receiver operating characteristics (ROC) analysis. In the speckle-free feature set, the area under the ROC curve was 0.96 for the support vector machines classifier whereas for the probabilistic neural networks was 0.91. In the feature set with speckle, the corresponding areas under the ROC curves were 0.88 and 0.86 respectively for the two classifiers. The proposed features can increase the classification accuracy and decrease the rate of missing and misdiagnosis in thyroid cancer control.

Mammographic mass classification using textural features and descriptive diagnostic data

Texture analysis is one of the most important factors in breast tissue characterization. An analytical approach to texture classification, combined with qualitative descriptive diagnostic data, is presented in this article. For qualitative data, a statistical approach was applied in detailed clinical findings and texture-related features were established as of most importance during the diagnostic assertion process. A complete set of textural feature functions in multiple configurations and implementations was applied to a large set of digitized mammograms, in order to establish the discriminating value and statistical correlation with qualitative texture descriptions of breast mass tissue. Multiple linear and non-linear models were applied during the classification process, including LDA, least-squares minimum distance, K-nearest-neighbors, RBF and MLP. Optimal classification accuracy rates reached 81.5% for texture-only classification and 85.4% with the introduction of patients age as an example of hybrid approaches.

Inter-scale wavelet analysis for speckle reduction in thyroid ultrasound images

A wavelet-based method for speckle suppression in ultrasound images of the thyroid gland is introduced. The classification of image pixels as speckle or part of important image structures is accomplished within the framework of back-propagation tracking and singularity detection of wavelet transform modulus maxima, derived from inter-scale analysis. A comparative study with other de-speckling techniques, employing quantitative indices, demonstrated that our method achieved superior speckle reduction performance and edge preservation properties. Moreover, a questionnaire regarding qualitative imaging parameters, emanating from various visual observations, was employed by two experienced physicians in order to evaluate the algorithms speckle suppression efficiency.

A fully automated complete segmentation scheme for mammograms

This paper presents a fully automated complete segmentation method for mammographic images. Image preprocessing techniques are first applied to mammograms to remove the noise and then a breast boundary extraction algorithm is implemented, in order to distinguish breast tissue from the background. Next, an improved version of an existing pectoral muscle scheme is performed and a new nipple segmentation technique is applied, detecting the nipple when it is in profile. This improves the estimated breast boundary and serves as a key-point for further processing of the image. This composite method has been implemented and applied to miniMIAS, one of the most well-known mammographic databases. This database consists of 322 mediolateral oblique (MLO) view mammograms, obtained via a digitization procedure. The results are evaluated by an expert radiologist and are very promising. Accordingly, it is expected that this procedure can produce improved results, when applied to high-quality digital mammograms.

Fuzzy C-means-driven FHCE contextual segmentation method for mammographic microcalcification detection

Abstract The frequency histogram of connected elements (FHCE) is a recently proposed algorithm that has successfully been applied in various medical image segmentation tasks. The FHCE is based on the idea that most pixels belong to the same class as their neighbouring pixels. However, the FHCE performance relies to a great extent on the optimal selection of a threshold parameter. Since evaluating segmentation results is a highly subjective process, a collection of threshold values must typically be examined. No algorithm has been proposed to automate the determination of the threshold parameter value of the FHCE. This study presents a method based on the fuzzy C-means clustering algorithm, designed to automatically generate optimal threshold values for the FHCE. This new approach was applied as a part of a structured sequence of image processing steps in order to facilitate segmentation of microcalcifications in digitized mammograms. A unique threshold value was generated for each mammogram, taking into account the different grey-level patterns based on different compositions of various breast tissues in it. The segmentation algorithm was tested on 100 mammograms (50 collected from the Mammographic Image Analysis Society and 50 normal mammograms onto which a number of simulated microcalcifications were generated). The algorithm was able to detect subtle microcalcifications with sensitivity ranging from 93 to 98%, False alarm ratio from 3 to 5% and false negatives variability from 2 to 3%.

Development of the probabilistic neural network-cubic least squares mapping (PNN-LSM 3 ) classifier to assess carotid plaque's risk

An efficient classification algorithm based on the cubic least squares mapping (LSM3) and the probabilistic neural network (PNN) classifier is proposed for assessing the carotid plaques risk of causing brain infarcts. Ultrasound images of 24 high-risk and 32 low-risk carotid plaques were manually segmented by an experienced physician using a custom developed software. Three textural features, related to the plaques internal composition, the PNN, and the PNNLSM 3 classification algorithms were used to design a classification system. PNN classification accuracy was 92.9%, misdiagnosing one high-risk and three low-risk plaques while the PNN-LSM3 managed to classify all plaques correctly. The proposed system may be of value to patient management as a second opinion tool, after it is tested on more data in a clinical environment.

Multi-resolution morphological analysis and classification of mammographic masses using shape, spectral and wavelet features

This study constitutes a comprehensive signal analysis approach to the morphological characterization of mammographic mass shape. Three distinct areas of shape morphology were exploited for feature extraction. Specifically, the radial distance signal, the DFT spectrum envelope and the DWT decomposition with multiple wavelet function choices, were analyzed by seven curve feature functions, as carriers of significant discriminating information. Classification was conducted against the morphological shape type identification, as well as the verified clinical diagnosis, using optimized feature set selections and combinations by multivariate statistical significance analysis. All available datasets and configurations were applied to a wide range of linear and neural classifiers, including linear discriminant analysis, least-squares minimum distance, K-nearest neighbor, RBF and MLP neural networks. Neural classifiers outperformed linear equivalents in all cases, producing an overall accuracy of 72.3% for morphological shape type identification and 89.2% for clinical diagnosis identification.

Multimodality imaging and state-of-art GPU technology in discriminating benign from malignant breast lesions on real time decision support system

The aim of this study was to design a pattern recognition system for assisting the diagnosis of breast lesions, using image information from Ultrasound (US) and Digital Mammography (DM) imaging modalities. State-of-art computer technology was employed based on commercial Graphics Processing Unit (GPU) cards and parallel programming. An experienced radiologist outlined breast lesions on both US and DM images from 59 patients employing a custom designed computer software application. Textural features were extracted from each lesion and were used to design the pattern recognition system. Several classifiers were tested for highest performance in discriminating benign from malignant lesions. Classifiers were also combined into ensemble schemes for further improvement of the systems classification accuracy. Following the pattern recognition system optimization, the final system was designed employing the Probabilistic Neural Network classifier (PNN) on the GPU card (GeForce 580GTX) using CUDA programming framework and C++ programming language. The use of such state-of-art technology renders the system capable of redesigning itself on site once additional verified US and DM data are collected. Mixture of US and DM features optimized performance with over 90% accuracy in correctly classifying the lesions.