Khalil Chtourou

3D multimodal MRI brain glioma tumor and edema segmentation: a graph cut distribution matching approach.

This study investigates a fast distribution-matching, data-driven algorithm for 3D multimodal MRI brain glioma tumor and edema segmentation in different modalities. We learn non-parametric model distributions which characterize the normal regions in the current data. Then, we state our segmentation problems as the optimization of several cost functions of the same form, each containing two terms: (i) a distribution matching prior, which evaluates a global similarity between distributions, and (ii) a smoothness prior to avoid the occurrence of small, isolated regions in the solution. Obtained following recent bound-relaxation results, the optima of the cost functions yield the complement of the tumor region or edema region in nearly real-time. Based on global rather than pixel wise information, the proposed algorithm does not require an external learning from a large, manually-segmented training set, as is the case of the existing methods. Therefore, the ensuing results are independent of the choice of a training set. Quantitative evaluations over the publicly available training and testing data set from the MICCAI multimodal brain tumor segmentation challenge (BraTS 2012) demonstrated that our algorithm yields a highly competitive performance for complete edema and tumor segmentation, among nine existing competing methods, with an interesting computing execution time (less than 0.5s per image).

Speckle noise reduction in breast ultrasound images: SMU (SRAD median unsharp) approch

In medical image processing, image denoising has become a very essential for better information extraction from the image and mainly from so noised ones, such as ultrasound (US)images. On the other hand, processed image must preserve the pertinent details of the primary image. So, arbitration between the perpetuation of useful diagnostic information and noise suppression must be teasured in medical images. In general we rely on the intervention of a proficient to control the quality of processed images. In certain cases, for instance in ultrasound images, the noise can restrain information which is valuable for the general practitioner. Consequently medical images are very inconsistent, and it is crucial to operate case to case. This paper presents a novel algorithm SMU (Srad Median Unsharp) for noise suppression in ultrasound breat images inorder to realize a computer aided diagnosis (CAD) for breast cancer. A comparitive study of the results obtained by the proposed method with the results achieved from the other speckle noise reduction techniques demonstrates its higher performance for speckle reduction.

Multi-slices breast ultrasound lesion segmentation using Multi-Scale Vector Field Convolution snake

This study aims to apply a novel method called Multi-scale Vector Field Convolution Snake (MVFC) to segment breast ultrasound images using all slices presenting the lesion. The key idea is to combine the Vector Field Convolution Snake (VFC) method with a two-dimensional Gaussian filter with variable standard deviations in order to make snake models less sensitive to speckle noise and to contrast quality. Experimental results show that the form of the lesion changes from one slice to another which allows achieving greater precision in the extraction of the lesion characteristics.

Towards factor analysis exploration applied to positron emission tomography functional imaging for breast cancer characterization

The paper aims to explore the factor analysis when applied to a dynamic sequence of medical images obtained using nuclear imaging modality, Positron Emission Tomography (PET). This latter modality allows obtaining information on physiological phenomena, through the examination of radiotracer evolution during time. Factor analysis of dynamic medical images sequence (FADMIS) estimates the unerlying fundamental spatial distributions by factor images and the associated so-called fundamental functions (describing the signal variations) by factors. This method is based on an orthogonal analysis followed by an oblique analysis. The results of the FADMIS are physiological curves showing the evolution during time of radiotracer within homogeneous tissues distributions. This functional analysis of dynamic nuclear medical images is considered to be very efficient for cancer characterization, vascularization as well as possible evaluation of response to therapy.

CT scan contrast enhancement using singular value decomposition and adaptive gamma correction

We propose in this paper a new enhancement algorithm dedicated to dark computed tomography (CT) scan based on discrete wavelet transform with singular value decomposition (DWT–SVD) followed by adaptive gamma correction (AGC). Discrete wavelet transform (DWT) is considered to decompose the input dark CT image in four sub-bands. Singular value decomposition (SVD) is used in order to compute the corresponding singular value matrix of low–low (LL) sub-band image. The enhanced LL sub-band is determined by scaling the singular value matrix of original LL sub-band by an adequate correction factor, followed by inverse SVD. For a further contrast improvement, the new enhanced LL sub-band image is processed using an AGC algorithm. Finally, the obtained LL sub-band image undergoes inverse DWT together with the unprocessed sub-bands to generate the final enhanced image. This proposed method has the advantage of being fully automatic and could be applied for dark input images with either low or moderate contrast. Different dark CT images are considered to compare the performance of our proposed method to three other enhancement techniques using both objective and subjective assessments. Simulation results show that our proposed algorithm consistently produces good contrast enhancement, with best brightness and edges details conservation and with minimum added distortions to the enhanced CT images.

EEG localization error exploration vs distance between sources

The study of the cerebral electric activity in the brain, requires estimation of the EEG inverse problem. Several methods of the inverse problem solution are studied in the literature. In this paper we propose a comparative study of the localization error based on distance between active sources, using WMN-FOCUSS, LORETA-FOCUSS, and Shrinking sLORETA-FOCUSS as inverse problem solution. So, we consider the presence of two active sources whose distance intervals range from 1cm to 10cm. This study is made in a first case using electrode configuration on the entire brain, and in a second case on each lobe separately. The presented results show that the new approach for the separately lobes gives a good localization of the active sources in the brain.