Norhene Gargouri Ben Ayed

LBPV descriptors-based automatic ACR/BIRADS classification approach

Mammogram tissue density has been found to be a strong indicator for breast cancer risk. Efforts in computer vision of breast parenchymal pattern have been made in order to improve the diagnostic accuracy by radiologists. Motivated by recent results in mammogram tissue density classification, a novel methodology for automatic American College of Radiology Breast Imaging Reporting and Data System classification using local binary pattern variance descriptor is presented in this article. The proposed approach characterizes the local density in different types of breast tissue patterns information into the LBP histogram. The performance of macro-calcification detection methods is developed using FARABI database. Performance results are given in terms of receiver operating characteristic. The area under curve of the corresponding approach has been found to be 79%.

A new human identification based on fusion fingerprints and faces biometrics using LBP and GWN descriptors

Single modality biometric recognition system is often not able to meet the desired system performance requirements. Several studies have shown that multimodal biometric identification systems improve the recognition accuracy and allow performances that are required for many security applications. In this paper, we have developed a multimodal biometric recognition system which combines two modalities: face and fingerprint. For face trait, we build features based on Gabor Wavelet Networks (GWNs), while Local Binary Patterns (LBP) is used for fingerprint trait. Experimental results affirm that a weighted sum based fusion achieves excellent recognition performances, which out performs both single biometric systems.

Diagnosis of masses in mammographic images based on Zernike moments and local binary attributes

Masses are important elements in the diagnosis of breast cancer. Many studies discussed the problem of detection and/or diagnosis of masses and most of these researches were based on shape descriptors to make decision. Textural descriptors contribute in indicating the presence of masses. Morphological descriptors determine their malignancy degree. Thus, we decided in our work to make a combination of morphological and textural descriptors. In fact, this method allowed us to extract different features in order to help make a decision concerning the malignancy of masses. The shape descriptor “Zernike moments” has the advantages to be invariant to the rotation and to be orthogonal. In addition, the texture descriptor “local binary attributes” provides information about the local variations of gray levels in the image. A multi-layer perceptron is used in the classification stage. The results were validated by using 160 regions of interest which are extracted from the database of mammographic images DDSM (Digital Database for Screening Mammography). We obtained an area under the ROC (Receiver Operating Characteristics) curve which is equal to 0,96. The results were confirmed by a radiologist.

FDDSM: Toward a full field digital mammographic database

Computer-Aided Detection and Diagnosis (CADD) systems have been created in the last two decades to help radiologists either in the automatic detection or diagnosis of abnormalities in mammographic images. Accordingly, the developed algorithms in an existing amount of mammographic images should be assessed on a large database. For a fair validation, the database should be very representative including different subjects and anomalies. Nevertheless, available mammographic images databases do not provide the required resolution quality and are not well covering all kinds of diseases such as architectural distortion and kists. They also lack some important subject details such as biopsy results, hereditary factors, etc..., thus making critical the generalization of research results. In this paper, we present a new mammographic image database Farabi Digital Database of Screening Mammography (FDDSM), collected in El Farabi radiology center. The digitization of mammographic images was made with high contrast and a resolution of 12 bits per pixel. The database covers 2052 mammograms for 342 subjects. It was collected on a three year time period. As illustration, the presented database was used, in previous work, for mass detection and classification and achieves good generalization performances.

An efficient microcalcifications detection based on dual spatial/spectral processing

Microcalcifications are tiny deposits of calcium located in breast tissue. They appeared as very small highlighted regions in comparison with their surrounding tissue. Spatial non linear enhancement can be applied for microcalcification detection. However, efficiency of a such approach depends on breast density: in case of extreme breast density, the contrast between microcalcification’s details and their surrounding tissue is attenuated leading to a limitation of spatially based approaches. In that case, frequency analysis such as wavelet based analysis can be more relevant for dissociating microcalcifications. The main goal of Computer Aided Detection systems (CAD) is to detect breast cancer at an early stage for all breast density classes by using entropies to enhance and then detect microcalcification details. Accordingly, we combine our approach a spatial Automatic Non Linear Stretching (ANLS) and Shannon Entropy based Wavelet Coefficient Thresholding (SE_WCT). Validation of the proposed approach is done on the Mammographic Image Analysis Society (MIAS) database. The evaluation of the contrast is based on the Second-Derivative-Like measure of enhancement(SDME). Accordingly, it yields to a mean SDME of 78.8dB on the whole database. The performance metric for evaluating our proposed CAD is the Receiver Operating Characteristic(ROC) curve and the free-response ROC (FROC). An area under the ROC curve Az = 0.92 is obtained as well as 97.14 % of True Positives (TP) with 0,48 False positives per image (FP).

A new Tsallis based automatic non linear enhancement of mammograms for microcalcifications segmentation in high density breast

Microcalcifications are tiny deposits of calcium located in breast tissue. They appeared as very small highlighted regions in comparaison with their surrounding tissue. The difference of contrast between microcalcifications and the normal tissue depend on the breast density: The more the breast is dense, the less is the contrast. In this context, we propose to enhance microcalcifications details for each type of breast density using for methods. As we know that the BIRADS/ACR 4 contains dense breast, Thats why we have proposed to make the Non Linear Stratching (NLS)automatic by applying an improved Tsallis entropy. The proposed mammography enhancement approach is evaluated on the Digital Database for Screening Mammography (DDSM) database.

New developments in the diagnostic procedures to reduce prospective biopsies breast

This paper studies the computer-aided diagnosis technique potential in discriminating accurately benign masses among a given subset of 100 patients which makes it possible to degrade cases from Breast Imaging-Reporting and Data System (BIRADS) 3 to BIRADS 2 avoiding prospective biopsies. Such accuracy is required since expert radiologists assign BIRADS3 category by default mostly for reducing false negative cases. We aim here at classifying masses on a risk rate scale for malignancy. The proposed system segments automatically potential masses and quantifies critical related features. A decision tree was accordingly applied. In a first level, a mass detection is based on a new local pattern model named Weighted Gray Level and Local Difference features (WGLLD) and a nearest neighborhood (NN) a classifier. In the second level, Zernike moment features were used for shape characterization with connection by an Artificial Neural Network (ANN) based classifier, after that we segment masses and extract shape features using Zernike moments. For validation purposes, a total of 100 lesions from local breast database (FDDSM)is used. Most of these cases are biopsy confirmed. The system successfully downgraded 7 cases over 41 rated by the expert as belonging to BIRADS 3 to BIRADS 2, but, it recommended biopsy for 41/100 atypical lesions. Ultimately, the system identified 59 benign lesions to BIRADS 2, 7 cases from these were classified as belonging to BIRADS 3 by the expert, and thus reached a reduction of unnecessary breast biopsies. The proposed CAD system allows a classification rate of 98% (only one benign case is missed). The proposed Computer Aided Diagnosis (CAD) system demonstrated the ability to predict benignancy of the most difficult cases.30 Appearance changes were also shown to be more characterizing after mammogram enhancement. With further validation, these results could form a substrate for a clinically useful computer-aided diagnosis tool which could provide earlier detection of breast cancer signs.

An efficient technique for the extraction of microcalcification's severity features

Microcalcifications are very tiny deposits of calcium allocated in the breast tissue. Their gray level is similar to the dense normal breast tissue so its very difficult to differentiate between them. Once detected, its very difficult to between malign end benign microcalcifications. In this paper, we apply a new method to extract features of microcalcifications in order to classify them into malign and benign. This technique, called the Discriminative Completed Local Binary Pattern (DisCLBP), extracts texture characteristics of breast tissue in order to characterize the severity of microcalcifications. Classification of these structures is accomplished through Artificial Neural Network (ANN), which separate them in two groups: malignant and benign microcalcifications. Performance results are given in terms of receiver operating characteristic (ROC). The area under curve (AUC) of the corresponding approach has been found to be 93.45%.

Multi-scale Gray Level and Local Difference for texture classification

In this paper, we present a novel extension of the Gray Level and Local Difference (GLLD) method and it is named as Multi-scale GLLD for texture classification. In the GLLD, a local region is described by its central pixel and the local difference sign-magnitude. The central pixels representing the image gray level are transformed into a binary code by global thresholding. The local difference sign-magnitude is based on the image decomposition into two complementary components: the signs and the magnitudes. By combining SGLLD, MGLLD, and CGLLD features, momentous improvement can be made in terms of texture classification. As an extension of the GLLD, we proposed to apply the multi-scale scheme and we obtained better results. The classification rate of the corresponding approach reached 96%. A comparative study with previous approaches confirms that the proposed approach presents the best performances.

A new medical decision system to identify breast cancer

An improved computer system has been presented to classify the mass and identify the different stages of breast cancer using artificial neural network (ANN). In this paper, we extract texture and shape features. The accuracy of the proposed system is 99.5%. The images from Farabi digital database for screening mammography have been applied for the development of the proposed system. This later may provide precious information to radiologists.