Fabián Narváez

Automatic BI-RADS description of mammographic masses

This paper presents a CBIR (Content Based Information Retrieval) framework for automatic description of mammographic masses according to the well known BI-RADS lexicon Unlike other approaches, we do not attempt to segment masses but instead, we describe the regions an expert selects, after the series of rules defined in the BI-RADS lexicon The content based retrieval strategy searches similar regions by automatically computing the Mahalanobis distance of feature vectors that describe main shape and texture characteristics of the selected regions A description of a test region is based on the BI-RADS description associated to the retrieved regions The strategy was assessed in a set of 444 masses with different shapes and margins Suggested descriptions were compared with a ground truth already provided by the data base, showing a precision rate of 82.6% for the retrieval task and a sensitivity rate of 80% for the annotation task.

A content-based retrieval of mammographic masses using the curvelet descriptor

Computer-aided diagnosis (CAD) that uses content based image retrieval (CBIR) strategies has became an important research area. This paper presents a retrieval strategy that automatically recovers mammography masses from a virtual repository of mammographies. Unlike other approaches, we do not attempt to segment masses but instead we characterize the regions previously selected by an expert. These regions are firstly curvelet transformed and further characterized by approximating the marginal curvelet subband distribution with a generalized gaussian density (GGD). The content based retrieval strategy searches similar regions in a database using the Kullback-Leibler divergence as the similarity measure between distributions. The effectiveness of the proposed descriptor was assessed by comparing the automatically assigned label with a ground truth available in the DDSM database.1 A total of 380 masses with different shapes, sizes and margins were used for evaluation, resulting in a mean average precision rate of 89.3% and recall rate of 75.2% for the retrieval task.

Multi-view information fusion for automatic BI-RADS description of mammographic masses

Most CBIR-based CAD systems (Content Based Image Retrieval systems for Computer Aided Diagnosis) identify lesions that are eventually relevant. These systems base their analysis upon a single independent view. This article presents a CBIR framework which automatically describes mammographic masses with the BI-RADS lexicon, fusing information from the two mammographic views. After an expert selects a Region of Interest (RoI) at the two views, a CBIR strategy searches similar masses in the database by automatically computing the Mahalanobis distance between shape and texture feature vectors of the mammography. The strategy was assessed in a set of 400 cases, for which the suggested descriptions were compared with the ground truth provided by the data base. Two information fusion strategies were evaluated, allowing a retrieval precision rate of 89.6% in the best scheme. Likewise, the best performance obtained for shape, margin and pathology description, using a ROC methodology, was reported as AUC = 0.86, AUC = 0.72 and AUC = 0.85, respectively.

An open access thyroid ultrasound image database

Computer aided diagnosis systems (CAD) have been developed to assist radiologists in the detection and diagnosis of abnormalities and a large number of pattern recognition techniques have been proposed to obtain a second opinion. Most of these strategies have been evaluated using different datasets making their performance incomparable. In this work, an open access database of thyroid ultrasound images is presented. The dataset consists of a set of B-mode Ultrasound images, including a complete annotation and diagnostic description of suspicious thyroid lesions by expert radiologists. Several types of lesions as thyroiditis, cystic nodules, adenomas and thyroid cancers were included while an accurate lesion delineation is provided in XML format. The diagnostic description of malignant lesions was confirmed by biopsy. The proposed new database is expected to be a resource for the community to assess different CAD systems.

Breast mass classification using orthogonal moments

Automatic classification of breast masses in mammograms has been considered a major challenge. Mass shape, margin and density define the malignancy level according to a standardized description, the BI-RADS lexicon. Unlike other approaches, we do not segment masses but instead, we attempt to describe entire regions. In this paper, continuos (Zernike) and discrete (Krawtchouk) orthogonal moments were used to characterize breast masses and their discriminant power to classify benign and malign masses, was assessed. Firstly, Regions of Interest selected by an expert are projected onto two sets of orthogonal polynomials functions, continuous and discrete, thereby drawing shape global information onto a feature space. Using a simple euclidean metric between vectors, the projected images are automatically classified as benign or malign by a k-nearest neighbor strategy. The parameter space is characterized using a set of 150 benign and 150 malign images. The whole method was assessed in a set of 100 masses with different shape and margins and the classification results were compared against a ground truth, already provided by the database. These results showed that discrete Krawtchouk outperformed Zernike moments, reaching an accuracy rate of 90,2% (compared to 81% for Zernike moments), while the area under the curve in a ROC evaluation yielded Az=0.93 and Az=0.85 for the Krawtchouk and Zernike strategies, respectively.

Kushkalla: A Web-Based Platform to Improve Functional Movement Rehabilitation

Telerehabilitation is a growing alternative to traditional face-to-face therapy, which uses technological solutions to cover rehabilitation care in both clinical centers and in-home programs. However, the current telerehabilitation systems are limited to deliver a set of exercise programs for some specific locomotor disability, without including tools that allow a quantitative analysis of the rehabilitation progress, in real-time, as well as the medical condition of patients. This paper presents the design and development of a novel web-based platform, named “Kushkalla”, that allows to perform movement assessment for creating personalized home-based therapy routines, integrating hardware and software tools for a quantitative analysis of locomotor movements based on motion capture, preprocessing, monitoring, visualization, storage and analysis, in real-time. The platform combines two motion capture strategies, the Kinect-based and IMU-based motion capture. In addition, a set of 2D and 3D graphical models, virtual environments, based on WebGL technology, and videoconference module are included to allow the interaction between user and clinician for enhancing the capability of the clinician to direct rehabilitation therapies.

Characterizing Architectural Distortion in Mammograms by Linear Saliency

Architectural distortion (AD) is a common cause of false-negatives in mammograms. This lesion usually consists of a central retraction of the connective tissue and a spiculated pattern radiating from it. This pattern is difficult to detect due the complex superposition of breast tissue. This paper presents a novel AD characterization by representing the linear saliency in mammography Regions of Interest (ROI) as a graph composed of nodes corresponding to locations along the ROI boundary and edges with a weight proportional to the line intensity integrals along the path connecting any pair of nodes. A set of eigenvectors from the adjacency matrix is then used to extract discriminant coefficients that represent those nodes with higher salient lines. A dimensionality reduction is further accomplished by selecting the pair of nodes with major contribution for each of the computed eigenvectors. The set of main salient lines is then assembled as a feature vector that inputs a conventional Support Vector Machine (SVM). Experimental results with two benchmark databases, the mini-MIAS and DDSM databases, demonstrate that the proposed linear saliency domain method (LSD) performs well in terms of accuracy. The approach was evaluated with a set of 246 RoI extracted from the DDSM (123 normal tissues and 123 AD) and a set of 38 ROI from the mini-MIAS collections (19 normal tissues and 19 AD) respectively. The classification results showed respectively for both databases an accuracy rate of 89 % and 87 %, a sensitivity rate of 85 % and 95 %, and a specificity rate of 93 % and 84 %. Likewise, the area under curve (Az) of the Receiver Operating Characteristic (ROC) curve was 0.93 for both databases.

Characterization of architectural distortion on mammograms using a linear energy detector

Architectural distortion is a breast cancer sign, characterized by spiculated patterns that define the disease malignancy level. In this paper, the radial spiculae of a typical architectural distortion were characterized by a new strategy. Firstly, previously selected Regions of Interest are divided into a set of parallel and disjoint bands (4 pixels the ROI length), from which intensity integrals (coefficients) are calculated. This partition is rotated every two degrees, searching in the phase plane the characteristic radial spiculation. Then, these coefficients are used to construct a fully connected graph whose edges correspond to the integral values or coefficients and the nodes to x and y image positions. A centrality measure like the first eigenvector is used to extract a set of discriminant coefficients that represent the locations with higher linear energy. Finally, the approach is trained using a set of 24 Regions of Interest obtained from the MIAS database, namely, 12 Architectural Distortions and 12 controls. The first eigenvector is then used as input to a conventional Support Vector Machine classifier whose optimal parameters were obtained by a leave-one-out cross validation. The whole method was assessed in a set of 12 RoIs with different distribution of breast tissues (normal and abnormal), and the classification results were compared against a ground truth, already provided by the data base, showing a precision rate of 0.583%, a sensitivity rate of 0.833% and a specificity rate of 0.333%.