Walid Barhoumi

Breast cancer diagnosis in digitized mammograms using curvelet moments

BACKGROUND Feature extraction is a key issue in designing a computer aided diagnosis system. Recent researches on breast cancer diagnosis have reported the effectiveness of multiscale transforms (wavelets and curvelets) for mammogram analysis and have shown the superiority of curvelet transform. However, the curse of dimensionality problem arises when using the curvelet coefficients and therefore a reduction method is required to extract a reduced set of discriminative features. METHODS This paper deals with this problem and proposes a feature extraction method based on curvelet transform and moment theory for mammogram description. First, we performed discrete curvelet transform and we computed the four first-order moments from curvelet coefficients distribution. Hence, two feature sets can be obtained: moments from each band and moments from each level. In this work, both sets are studied. Then, the t-test ranking technique was applied to select the best features from each set. Finally, a k-nearest neighbor classifier was used to distinguish between normal and abnormal breast tissues and to classify tumors as malignant or benign. Experiments were performed on 252 mammograms from the Mammographic Image Analysis Society (mini-MIAS) database using the leave-one-out cross validation as well as on 11553 mammograms from the Digital Database for Screening Mammography (DDSM) database using 2×5-fold cross validation. RESULTS Experimental results prove the effectiveness and the superiority of curvelet moments for mammogram analysis. Indeed, results on the mini-MIAS database show that curvelet moments yield an accuracy of 91.27% (resp. 81.35 %) with 10 (resp. 8) features for abnormality (resp. malignancy) detection. In addition, empirical comparisons of the proposed method against state-of-the-art curvelet-based methods on the DDSM database show that the suggested method does not only lead to a more reduced feature set, but it also statistically outperforms all the compared methods in terms of accuracy. CONCLUSIONS In summary, curvelet moments are an efficient and effective way to extract a reduced set of discriminative features for breast cancer diagnosis.

An efficient image-mosaicing method based on multifeature matching

Mosaicing is connecting two or more images and making a new wide area image with no visible seam-lines. Several algorithms have been proposed to construct mosaics from image sequence where the camera motion is more or less complex. Most of these methods are based either on the interest points matching or on theoretical corner models. This paper describes a fully automated image-mosaicing method based on the regions and the Harris points primitives. Indeed, in order to limit the search window of potential homologous points, for each point of interest, regions segmentation and matching steps are being performed. This enables us to improve the reliability and the robustness of the Harris points matching process by estimating the camera motion. The main originality of the proposed system resides in the preliminary manipulation of regions matching, thus making it possible to estimate the rotation, the translation and the scale factor between two successive images of the input sequence. This estimation allows an initial alignment of the images along with the framing of the interest points search window, and therefore reducing considerably the complexity of the interest points matching algorithm. Then, the resolution of a minimization problem, altogether considering the couples of matched-points, permits us to perform the homography. In order to improve the mosaic continuity around junctions, radiometric corrections are applied. The validity of the herewith described method is illustrated by being tested on several sequences of complex and challenging images captured from real-world indoor and outdoor scenes. These simulations proved the validity of the proposed method against camera motions, illumination variations, acquirement conditions, moving objects and image noise. To determine the importance of the regions matching stage in motion estimation, as well as for the framing of the search window associated to a point of interest, we compared the matching points results of this described method with those produced using the zero-mean normalized cross correlation score (without regions matching). We made this comparison in the case of a simple motion (without the presence of a rotation around optical axis and/or a scale factor), in the case of a rotation and in the general case of an homothety. For justifying the effectiveness of this method, we proposed an objective assessment by defining a reconstruction error.

A robust framework for joint background/foreground segmentation of complex video scenes filmed with freely moving camera

This paper explores a robust region-based general framework for discriminating between background and foreground objects within a complex video sequence. The proposed framework works under difficult conditions such as dynamic background and nominally moving camera. The originality of this work lies essentially in our use of the semantic information provided by the regions while simultaneously identifying novel objects (foreground) and non-novel ones (background). The information of background regions is exploited to make moving objects detection more efficient, and vice-versa. In fact, an initial panoramic background is modeled using region-based mosaicing in order to be sufficiently robust to noise from lighting effects and shadowing by foreground objects. After the elimination of the camera movement using motion compensation, the resulting panoramic image should essentially contain the background and the ghost-like traces of the moving objects. Then, while comparing the panoramic image of the background with the individual frames, a simple median-based background subtraction permits a rough identification of foreground objects. Joint background-foreground validation, based on region segmentation, is then used for a further examination of individual foreground pixels intended to eliminate false positives and to localize shadow effects. Thus, we first obtain a foreground mask from a slow-adapting algorithm, and then validate foreground pixels (moving visual objects + shadows) by a simple moving object model built by using both background and foreground regions. The tests realized on various well-known challenging real videos (across a variety of domains) show clearly the robustness of the suggested solution. This solution, which is relatively computationally inexpensive, can be used under difficult conditions such as dynamic background, nominally moving camera and shadows. In addition to the visual evaluation, spatial-based evaluation statistics, given hand-labeled ground truth, has been used as a performance measure of moving visual objects detection.

Effective Region-based Relevance Feedback for Interactive Content-based Image Retrieval

This paper proposes an effective framework for interactive region-based image retrieval. By utilizing fuzzy coarse segmentation and the graph structure for representing each image, the retrieval process was performed by measuring the image similarity according to the graph similarity. To assess the similarity between two graphs, fuzzy inter relations among regions feature vectors and spatial dispositions as well as fuzzy regions weights are explored. A region-based relevance feedback scheme was also incorporated into the retrieval process, by updating the importance of query image regions based on the user feedbacks, leading to a further performance improvement. Experimental study proves that the proposed region-based relevance feedback mechanism tailors the system semantic behavior relatively to each user personal preferences through the accumulation of the useful semantic information from the feedback information.

Unsupervised background reconstruction based on iterative median blending and spatial segmentation

We propose in this paper a novel iterative approach for unsupervised reconstruction of static background from a complex video shot. After aligning some key-frames of the video onto a reference plane in order to compensate the camera motion, the basic idea of the suggested scheme is to iteratively reconstruct a precise image of the background using median blending and spatial segmentation. In each iteration, coarse binary masks, representing foreground moving objects, are estimated by comparing each motion-compensated key-frame with the corresponding part in the input background image. These masks are then refined by spatial segmentation while profiting of the semantic information offered by region maps. The iterative process allows the blending operator to eliminate the detected moving objects while reconstructing the output background image. Several experiments have been carried out to prove the effectiveness of the suggested unsupervised approach for precise background reconstruction of complex dynamic scenes after a relatively small number of iterations.

Video Similarity Measurement Based on Attributed Relational Graph Matching

In this paper, an original scheme for video similarity detection is proposed in order to establish correspondence between two video sequences. This scheme consists first to summarize the visual contents of a video sequence in a small set of images. Each image is then modeled, by an Attributed Relational Graph (ARG), as the composition of salient objects with specific spatial relationship. Matching two video sequences is thereby reduced to the ARG similarity problem. The proposed approach offers a principled way to define the ARG similarity that accounts for both the attribute and topological differences of the two considered ARGs. Indeed, we proposed herein a cost-efficient solution to find the best alignment between two ARGs. This consists to the minimization of a similarity measure between the two graphs using dynamic programming. This measure can be considered as a matching rate which can be very useful for Content Based Video Retrieval (CBVR) applications. The suggested scheme was preliminary tested on real-world databases and very promising results were observed.

A Collaborative System for Pigmented Skin Lesions Malignancy Tracking

This paper presents a new collaborative computer-aided diagnosis system for skin lesions malignancy tracking composed of two modules for lesion description and decision. The decision module incorporates classification and content-based image retrieval schemes (CBIR). The final decision of lesion malignancy will be obtained by merging the two decisions while using the Dempster-Shafer theory in order to improve the accuracy of the final produced decisions. Indeed, after the preprocessing of the studied image and the extraction of the skin lesions by the segmentation process, the lesion description stage defines a set of descriptive features reflecting the clinical signs of the considered lesions malignancy. In fact, 21 features representing shape and radiometric properties are calculated. The quality of these features is evaluated by applying principal components analysis (PCA) and ROC assessment criteria. The results show that the feature set can be reduced to dimension 16. Then, the proposed system estimates the preliminary lesions class with the classification scheme, while using a perceptron neural network technique preceded by a training step. Moreover, given a database of skin lesions, CBIR of the images belonging to this database which gather with the studied image and whose lesions malignancy states are known, permits to have another preliminary idea on the type of the eventual skin lesion. Finally, the results of classification and retrieval schemes are combined while using the Dempster-Shafer theory. This consists to consider the results produced separately by each technique as being dubious sources of information on the lesions malignity with an aim of combining their respective opinions. The proposed architecture allows the production of a viable cost-effective set of opinions on skin lesions malignancy. Besides, since the decision can never be perfect, the CBIR subsystem displays also visually similar images with known pathologies, to provide an intuitive aid to the dermatologist to improve the diagnosis accuracy.

Multi-view score fusion for content-based mammogram retrieval

Screening mammography provides two views for each breast: Medio-Lateral Oblique (MLO) and Cranial-Caudal (CC) views. However, current content based image retrieval (CBIR) systems analyze each view independently, in spite of their complementarities. To further improve the retrieval performance, this paper introduces a two-view CBIR system that combines retrieval results of MLO and CC views. First, we computed the similarity scores between MLO (resp. CC) ROIs in the database and the MLO (resp. CC) query ROI. These ROIs are characterized using curvelet moments. Then, a new linear weighted sum scheme combines MLO and CC scores; it assigns weights for each view according to the distribution of the classes of its neighbors. The ROIs having the highest fused scores are displayed to the radiologist and used to compute the malignancy likelihood of the lesion. Experiments performed on mammograms from the Digital Database for Screening Mammography (DDSM) show the effectiveness of the proposed method.

Towards a standard approach for medical images segmentation

Summary form only given. In this paper we introduce a standard approach for medical images segmentation. In fact, for these images the segmentation consists in the extraction of an area of interest representing the organ subject of diagnosis. We distinguish two approaches depending on whether this area is composed of one region or of many regions. If it is composed of a single region, we introduce a growing region algorithm after a prestep based on fuzzy sets. Otherwise, we introduce an approach integrating a hierarchical system of segmentation in regions and a system of fuzzy classification. We illustrate our approach by applying it on real images in the frameworks of dermatology, neurology and mammography.

Objective and cost-efficient approach for skin lesions classification

Summary form only given. We deal with developing methods for an objective and cost-efficient tool for diagnosing skin lesions based on digitized dermatoscopic color images. We define a segmentation approach by fuzzy region growing applied to the Karhunen-Loeve transform of the RGB color vectors to separate pigmented lesion from the surrounding healthy skin. Then, a set of 14 characteristics of the lesion, represented by a set of numbers called feature scores, is extracted from the binary mask of the lesion deduced by the segmentation step. The quality of the features is evaluated by applying several feature selection methods in order to eliminate redundant information and accelerate the further classification step. Results show that most selection methods allow one to reduce the feature set to dimension five, which permits considerable reduction of calculation time, without significant loss of information. Feeding the selected features to a multilayer perception classifier allows us to generate a computerized diagnosis, suggesting whether the lesion is benign or malignant. With this approach, for reasonably balanced training/testing sets, we obtain above 77% correct classification of the malignant and benign tumors on real skin images.