Aymen Mouelhi

Automatic image segmentation of nuclear stained breast tissue sections using color active contour model and an improved watershed method

Abstract Automatic image segmentation of immunohistologically stained breast tissue sections helps pathologists to discover the cancer disease earlier. The detection of the real number of cancer nuclei in the image is a very tedious and time consuming task. Segmentation of cancer nuclei, especially touching nuclei, presents many difficulties to separate them by traditional segmentation algorithms. This paper presents a new automatic scheme to perform both classification of breast stained nuclei and segmentation of touching nuclei in order to get the total number of cancer nuclei in each class. Firstly, a modified geometric active contour model is used for multiple contour detection of positive and negative nuclear staining in the microscopic image. Secondly, a touching nuclei method based on watershed algorithm and concave vertex graph is proposed to perform accurate quantification of the different stains. Finally, benign nuclei are identified by their morphological features and they are removed automatically from the segmented image for positive cancer nuclei assessment. The proposed classification and segmentation schemes are tested on two datasets of breast cancer cell images containing different level of malignancy. The experimental results show the superiority of the proposed methods when compared with other existing classification and segmentation methods. On the complete image database, the segmentation accuracy in term of cancer nuclei number is over than 97%, reaching an improvement of 3–4% over earlier methods.

Linear feature selection and classification using PNN and SFAM neural networks for a nearly online diagnosis of bearing naturally progressing degradations

In this work, an effort is made to characterize seven bearing states depending on the energy entropy of Intrinsic Mode Functions (IMFs) resulted from the Empirical Modes Decomposition (EMD). Three run-to-failure bearing vibration signals representing different defects either degraded or different failing components (roller, inner race and outer race) with healthy state lead to seven bearing states under study. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are used for feature reduction. Then, six classification scenarios are processed via a Probabilistic Neural Network (PNN) and a Simplified Fuzzy Adaptive Resonance Theory Map (SFAM) neural network. In other words, the three extracted feature data bases (EMD, PCA and LDA features) are processed firstly with SFAM and secondly with a combination of PNN-SFAM. The computation of classification accuracy and scattering criterion for each scenario shows that the EMD-LDA-PNN-SFAM combination is the suitable strategy for online bearing fault diagnosis. The proposed methodology reveals better generalization capability compared to previous works and it is validated by an online bearing fault diagnosis. The proposed strategy can be applied for the decision making of several assets. Display Omitted A new methodology proposed for a nearly online damage stage detection.The proposed strategy is based on neural networks and EMD method.The nearly online detection of bearing health status is done thanks to various damage stages.Experimental results show that this methodology is very effective.Unlike previous works, the proposed method is tested for the diagnosis of naturally progressing bearing degradations.

Automatic segmentation of clustered breast cancer cells using watershed and concave vertex graph

Automatic segmentation of stained breast tissue images helps pathologists to discover the cancer disease earlier. Separation of touching cells presents many difficulties to the traditional segmentation algorithms. In this paper, we propose a new automatic method to segment clustered cancer cells. In the first step, we detect cell regions using a modified geometric active contour based on Chan-Vese energy functional. Then, touching cell regions are extracted from the pre-segmented image by detecting high concavity points along the cell contours. A gradient-weighted distance transform is used in the watershed algorithm in order to get the most significant inner edges. To solve the problem of over-segmentation, which is the major drawback of the watershed method, a combination of three techniques is presented as a post-processing step. First, the nearest end points to concave vertices are detected in the inner edges in order to get the initial separating curve candidates. Second, a concave vertex graph is constructed from the end points and the separating curves. Finally, Dijkstra algorithm is applied to find the shortest path that separates the touching cells. The proposed algorithm is tested on several breast cancer cell images and its compared with the classical watershed algorithm and a recent marker-controlled watershed method. The experimental results show the performance of the presented approach.

Accelerating the Multilayer Perceptron Learning with the Davidon Fletcher Powell Algorithm

In this paper, the Davidon Fletcher Powell (DFP) algorithm for nonlinear least squares is proposed to train multilayer perceptron (MLP). Applied on both a single output layer perceptron and MLP, we find that this algorithm is faster than the Marquardt-Levenberg (ML) algorithm known as the fastest algorithm used to train MLP until now. The number of iterations required by DFP algorithm to converge is less than about 50% of what is required by the ML algorithm. Interpretations of these results are provided in the paper.

Features extraction for medical characterization of nystagmus

Vertigo is a false sense of motion; the cause of this symptom is very complex and it needs a complementary exam. In this study, a vestibular diseases analysis method is proposed from videonystagmography (VNG) applications using an estimation of pupil movements in the case of uncontrolled motion to prove an efficient and reliable diagnosis results. Firstly, we applied a robust segmentation to extract the pupil position and radius based on pupil fitting and the center of ellipse using circular Hough transform to characterize the pupil movement. Secondly, we have computed the frequency variation of nystagmus in the fast and slow phases as additional features used in our VNG analysis. Finally, significant features are selected using a Fisher linear discriminant analysis. The obtained results are very interesting since we succeeded to remediate to the problems of VNG processing like bad performance caused by the eye blinking and the little information provided by the fast phase movement. We also succeeded to raise the most significant temporal features acting in vertigo. At last, one can say that the proposed study provide high requirements for diagnosis and analysis of vestibular dysfunction causing vertigo.

Application of feature reduction techniques for automatic bearing degradation assessment

Bearings are important assets for most industrial applications. The non-destructive diagnosis of these elements needs an accurate and reliable acquisition of its dynamic vibration signals affected by noise and the other part of system such as gears, shafts, etc. Empirical mode decomposition is an advanced signal processing tool for bearing fault feature extraction. In this paper, empirical mode decomposition is used to decompose non-linear and non-stationary bearing vibration signals into several stationary intrinsic mode functions and the empirical mode decomposition energy entropy is computed for each intrinsic mode function. Moreover, principal component analysis and linear discriminant analysis are used for feature reduction. Based on the Fishers criterion, experimental results show that linear discriminant analysis features are highlighted compared to principal component analysis features and original empirical mode decomposition features for bearing fault diagnosis as type (inner race, outer race, rolling element) and severity (normal, degraded, faulting).

A novel morphological segmentation method for evaluating estrogen receptors' status in breast tissue images

In this paper, we propose a fully automated method able to perform accurate nuclear segmentation in immunohistochemical breast tissue images in order to provide quantitative assessment of estrogen receptors status that will help pathologists in their diagnosis. The presented approach is based on color deconvolution and an enhanced morphological processing, which is used to identify positive stained nuclei and to separate all touching nuclei in the microscopic image for a subsequent cancer evaluation. Experiments on several breast cancer images of different patients admitted into the Tunisian Salah Azaiez Cancer Center, show the efficiency of the proposed method when compared to the manual evaluation of experts.

A new preprocessing parameter estimation based on geodesic active contour model for automatic vestibular neuritis diagnosis

The diagnostic of the vestibular neuritis (VN) presents many difficulties to traditional assessment methods This paper deals with a fully automatic VN diagnostic system based on nystagmus parameter estimation using a pupil detection algorithm. A geodesic active contour model is implemented to find an accurate segmentation region of the pupil. Hence, the novelty of the proposed algorithm is to speed up the standard segmentation by using a specific mask located on the region of interest. This allows a drastically computing time reduction and a great performance and accuracy of the obtained results. After using this fast segmentation algorithm, the obtained estimated parameters are represented in temporal and frequency settings. A useful principal component analysis (PCA) selection procedure is then applied to obtain a reduced number of estimated parameters which are used to train a multi neural network (MNN). Experimental results on 90 eye movement videos show the effectiveness and the accuracy of the proposed estimation algorithm versus previous work.

A novel automatic diagnostic approach based on nystagmus feature selection and neural network classification

Diagnosis of vertigo disorders presents many complications in the evaluation and treatment. In clinical practice, videonystagmography (VNG) tests are still an excellent bedside examination tool for vestibular disorder diagnosis. The parameters of different tests are used to get significant medical characterization of this disease. In this paper, we propose an approach to develop the assessment of vertigo symptom by the selection of the most pertinent VNG parameters using Fisher Linear Discriminant analysis. Therefore, a multilayer neural network (MNN) classifier is applied for automatic VNG dataset analysis based upon the fundamental measurements of normal and affected patients by vestibular disorder. The experimental results prove that the proposed approach is very interesting and helpful for an accurate diagnostic of this disease.

A supervised segmentation scheme based on multilayer neural network and color active contour model for breast cancer nuclei detection

Breast cancer nuclei detection is an impressive challenge in surgeries and medical treatments. In the microscopic image of immunohistologically stained breast tissue, cancer nuclei present a large variety in their characteristics that bring various difficulties for traditional segmentation algorithms. In this paper, we propose an efficient supervised segmentation method using a multilayer neural network (MNN) combined with a modified geometric active contour model based on Bayes error energy functional for nuclear stained breast tissue images. First, a discrimination function is constructed from color information of the desired nuclei using Fisher Linear Discriminant (FLD) analysis and a trained MNN in order to get a preliminary classification of cancer nuclei. This function is then included in the region term of the energy functional and the stopping function of the model to improve the segmentation accuracy of the detected cancer nuclei. Furthermore, the initial curve and the controlling parameters of the proposed model are estimated directly from the initial segmentation by the FLD-MNN method. The proposed segmentation scheme is tested on different microscopic breast tissue images recorded from real patients located in the Tunisian Salah Azaiez Cancer Center. The experimental results show the superiority of the proposed method when compared with other existing segmentation methods.