Rachid Jennane is a full Professor of image processing at the University of Orleans (France) where he is affiliated to Institut Denis Poisson (IDP - UMR CNRS 7013). He received the Ph. D. degree in electrical engineering from the University of Orleans (France). His Ph. D. concerned fractal modeling of textures with an application to bone microarchitecture analysis. He has been the principal investigator of several research projects in the image & signal processing areas. He supervised more than 20 PhD and Master Students in the area of signal & image processing. His current research interests include the processing of nD medical images in a broad sense, although his focus is on machine and deep learning. Since 2006, he receives an annual grant of excellence research from the French Government. He co-authored numerous original journals and proceeding articles. Pr. Jennane also spent the academic year 1998 as a visiting professor at the Electrical Engineering Department of the University of Rhode Island (USA). He is a reviewer for major conferences and journals in the field of image analysis and pattern recognition. In 2015, he chaired and organized the IEEE-IPTA Conference. In 2014, in conjunction and with the support of IEEE-ISBI, he organized the TCB-Challenge. Since 2012, he is a member of the different Committees of IEEE-IPTA Conferences. He coedited two Special Sections on Driven Visual Information and Advances in Computational Intelligence for Multimodal Biomedical Imaging. Phone: +33 2 38 49 45 38 Address: Pr. Rachid Jennane University of Orleans Bâtiment de Mathématiques Institut Denis Poisson - UMR CNRS 7013 Rue de Chartres BP 6759 45067 Orléans cedex 2 FRANCE
Trabecular bone (TB) characterization for osteoporosis diagnosis on plain radiographic images pre... more Trabecular bone (TB) characterization for osteoporosis diagnosis on plain radiographic images presents a challenging task in medical imaging. The goal of this paper is to study the information of complex wavelet coefficients to extract descriptors of the trabecular bone. These descriptors are extracted using some statistics of a new relative phase coefficients. The relative phase modeling is performed using two well-known circular models, called Von Mises and Wrapped Cauchy. Both models have the advantages of simple feature extraction using maximum likelihood estimators and similarity measurement of the Kullback-Leibler divergence, which is very helpful in the runtime task. Our proposed approach is used to recognize the osteoporosis patients from a population composed of both osteoporosis patients and control cases. Experimental results on a TB radiograph database show that considering the proposed approach improves the classification performances over the state-of-the-art methods.
This paper presents a new method for the characterization of trabecular bone texture variations f... more This paper presents a new method for the characterization of trabecular bone texture variations for early detection of osteoporosis. It relies on the study of texture variations in the complex-anisotropic domain associated with the Fully Anisotropic Morlet transform (FAM). Unlike conventional oriented wavelet-based schemes, this not only allows analyzing the texture over a wider range of scales and directions but also enables to consider the texture anisotropy. More specifically, we propose a new directional-signature which is a function of the orientation and relative magnitudes to characterize the radiographic bone anisotropy. The investigation of the inter-scale dependencies within an complex- anisotropic neighboring of FAM coefficients allows to capture separately the local-statistical properties of each directional sub-band, which is of great interest for the analysis of the trabecular bone texture exhibiting complex-irregular behavior. Results show that the proposed feature vector enables to discriminate two populations composed of Osteoporotic Patients (OP) and Control Cases (CC) with an Area Under Curve (AUC) rate of 91.28%.
Bone fractures caused by the osteoporosis become major problem of public health, and therefore, t... more Bone fractures caused by the osteoporosis become major problem of public health, and therefore, this subject becomes an increasingly important goal for both clinicians and biomedical researchers. The clinical implementation of three dimensional Computed Tomography Finite Element (3D CT/FE) methods is still limited due to the requirement of expensive computer hardware to achieve solutions of 3D FE models within a clinically acceptable time, as well as, the need for robust 3D segmentation and meshing techniques. Segmentation, meshing and FE analysis of a two dimensional (2D) geometry can be accomplished fast and are potentially more robust than of 3D CT/FE. The purpose of this study is to propose a 2D FE model derived from Dual-Energy X-ray Absorptiometry images (DXA), for possible clinical use with a high-quality compromise between the complexity and capability of the simulation. The results obtained suggest that 2D models can be applied with high accuracy to simulate the fracture failure and fracture type pattern. This model can provide a practical and rapid tool for which is capable in helping the clinical purposes.
Cette these est composee de deux parties. La premiere a pour but de valider une analyse fractale ... more Cette these est composee de deux parties. La premiere a pour but de valider une analyse fractale orientee (afo) de la texture d'une image ; la deuxieme est une application de cette analyse a des images radiographiques d'os. L'afo d'une image consiste a estimer le parametre h du mouvement brownien fractionnaire (fbm) le long de lignes de niveaux de gris paralleles dans des directions differentes. Le but est double: chiffrer la rugosite de la texture et rendre compte de son anisotropie. Pour valider cette methode sur des images synthetiques, nous proposons tout d'abord une methode pour evaluer la qualite des techniques de synthese 1d du fbm. Puis, sur des signaux de reference, nous comparons differents estimateurs de h pour ne retenir que les meilleurs en termes de biais et de variance. Les methodes de synthese selectionnees dans le cas 1d sont ensuite etendues pour obtenir des textures isotropes et anisotropes. Ces images sont ensuite analysees par afo et les resultats sont reportes sur un diagramme polaire dont l'analyse harmonique permet de valider la methode et d'extraire des parametres caracteristiques de la texture. L'afo est ensuite utilisee pour caracteriser la texture des images osseuses et quantifier le processus de decalcification de l'os. En premier lieu, les images d'os de deux populations (temoins, osteoporotiques) sont analysees. Ensuite, des resultats issus de l'afo sont interpretes en liaison avec ceux d'une analyse histomorphometrique et ceux d'une evaluation de la resistance osseuse par des tests biomecaniques. Finalement, d'autres attributs de texture sont estimes sur nos images d'os par matrice de co-occurence et matrice de longueurs de plages. Quoique perfectible, l'outil que nous proposons semble prometteur en vue de quantifier les changements architecturaux de l'os trabeculaire lies a l'osteoporose
HAL (Le Centre pour la Communication Scientifique Directe), 1996
International audienceThe resistance of bone tissue is influenced not only by bone density parame... more International audienceThe resistance of bone tissue is influenced not only by bone density parameters but also by bone architecture parameters, such as the microarchitecture and anisotropy of trabecular bone. We have developed and validated a fractal analysis method for studying bone microarchitecture on roentgenograms. This technique provides reproducible measurements of the fractal dimension (D) of bone, which reflects bone texture. The fractal dimension is determined in 36 different directions; the mean of these 36 values is representative of the image. A polar diagram gives the value of D according to the angle of analysis. By decomposing this diagram using polar Fourier Transform analysis, the parameters related to the shape of the polar diagram can be determined. This diagram image analysis technique has been used for other similar diagrams and applied to the results of our fractal analysis method. Diagram shape characterization may provide information on the angular distribution of results and therefore on the anisotropy of the images under study. The purpose of this study was to compare roentgenograms of the calcaneus and radius in the same subjects to determine whether texture and anisotropy parameters discriminated between these two bones. Roentgenograms of the calcaneus and radius were obtained in ten nonosteoporotic subjects. The radius had a smaller fractal dimension than the calcaneus (mean +/- standard deviation: 1.215 +/- 0.025 and 1.285 +/- 0.066, respectively; p = 0.014). Differences in the shape of the polar diagram were found between the two bones. The mean Fourier coefficient ratio C2/C4 was considerably smaller at the calcaneus (0.63 +/- 0.50) than at the radius (4.88 +/- 3.45; p = 0.005). Our method allows quantitative characterization of texture and anisotropy differences between the calcaneus and radius. The smaller fractal dimension of the radius probably reflects the simpler architecture of this non weight-bearing bone. The differences in polar diagram shape allow to evaluate anisotropy differences between the calcaneus and radius
HAL (Le Centre pour la Communication Scientifique Directe), 1999
International audienceThe overall goal of this project is to propose a non-invasive method for a ... more International audienceThe overall goal of this project is to propose a non-invasive method for a precise evaluation of bone architectural changes during osteoporosis. Our approach is based on a fractal analysis of the anisotropic texture of trabecular bone radiographs. Texture roughness is characterised by the fractional Brownian motion (FBM) model of parameter H. First, eight estimation methods for H were compared using reference FBM signals. It is shown that the maximum likelihood estimator (MLE) method provides the most accurate estimate on which to base a fractal analysis. Two parameters were derived: the first one quantified the average roughness of the texture while the other analysed roughness relative to its anisotropy. Subsequently, the image acquisition process used was characterised and optimised in order to ensure a satisfactory long-term reproducibility. Finally, the fractal analysis using MLE was applied to clinical data. It was found that the mean roughness was significantly different on osteoporotic and control cases. It was also demonstrated that mean roughness was also related to histomorphometric measures of the two-dimensional bone architecture. Finally, our research demonstrated that anisotropy parameter and bone strength parameters recovered after compression tests were correlated
2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), Mar 28, 2022
Knee OsteoArthritis (OA) is a common musculoskeletal disorder, which causes reduced mobility for ... more Knee OsteoArthritis (OA) is a common musculoskeletal disorder, which causes reduced mobility for seniors. Due to the semi-quantitative nature of the Kellgren-Lawrence (KL) grading system, medical practitioners’ grading is subjective, being entirely based on their experience. With the development of computer vision, Computer-Aided Diagnosis (CAD) systems based on deep learning methods such as convolutional neural networks (CNNs) have shown success in knee OA diagnosis. In this paper, we propose a new approach, the so-called Siamese-GAP Network, for the early detection of knee OA through a KL-grade classification. More precisely, a set of Global Average Pooling (GAP) layers is integrated into the Siamese network used to extract features from each level. The obtained features are then combined to improve the classification performance. Our experimental results on baseline X-ray images from the OsteoArthritis Initiative (OAI) dataset show that the proposed approach presents potential results for the early detection of knee OA.
Trabecular bone (TB) characterization for osteoporosis diagnosis on plain radiographic images pre... more Trabecular bone (TB) characterization for osteoporosis diagnosis on plain radiographic images presents a challenging task in medical imaging. The goal of this paper is to study the information of complex wavelet coefficients to extract descriptors of the trabecular bone. These descriptors are extracted using some statistics of a new relative phase coefficients. The relative phase modeling is performed using two well-known circular models, called Von Mises and Wrapped Cauchy. Both models have the advantages of simple feature extraction using maximum likelihood estimators and similarity measurement of the Kullback-Leibler divergence, which is very helpful in the runtime task. Our proposed approach is used to recognize the osteoporosis patients from a population composed of both osteoporosis patients and control cases. Experimental results on a TB radiograph database show that considering the proposed approach improves the classification performances over the state-of-the-art methods.
This paper presents a new method for the characterization of trabecular bone texture variations f... more This paper presents a new method for the characterization of trabecular bone texture variations for early detection of osteoporosis. It relies on the study of texture variations in the complex-anisotropic domain associated with the Fully Anisotropic Morlet transform (FAM). Unlike conventional oriented wavelet-based schemes, this not only allows analyzing the texture over a wider range of scales and directions but also enables to consider the texture anisotropy. More specifically, we propose a new directional-signature which is a function of the orientation and relative magnitudes to characterize the radiographic bone anisotropy. The investigation of the inter-scale dependencies within an complex- anisotropic neighboring of FAM coefficients allows to capture separately the local-statistical properties of each directional sub-band, which is of great interest for the analysis of the trabecular bone texture exhibiting complex-irregular behavior. Results show that the proposed feature vector enables to discriminate two populations composed of Osteoporotic Patients (OP) and Control Cases (CC) with an Area Under Curve (AUC) rate of 91.28%.
Bone fractures caused by the osteoporosis become major problem of public health, and therefore, t... more Bone fractures caused by the osteoporosis become major problem of public health, and therefore, this subject becomes an increasingly important goal for both clinicians and biomedical researchers. The clinical implementation of three dimensional Computed Tomography Finite Element (3D CT/FE) methods is still limited due to the requirement of expensive computer hardware to achieve solutions of 3D FE models within a clinically acceptable time, as well as, the need for robust 3D segmentation and meshing techniques. Segmentation, meshing and FE analysis of a two dimensional (2D) geometry can be accomplished fast and are potentially more robust than of 3D CT/FE. The purpose of this study is to propose a 2D FE model derived from Dual-Energy X-ray Absorptiometry images (DXA), for possible clinical use with a high-quality compromise between the complexity and capability of the simulation. The results obtained suggest that 2D models can be applied with high accuracy to simulate the fracture failure and fracture type pattern. This model can provide a practical and rapid tool for which is capable in helping the clinical purposes.
Cette these est composee de deux parties. La premiere a pour but de valider une analyse fractale ... more Cette these est composee de deux parties. La premiere a pour but de valider une analyse fractale orientee (afo) de la texture d'une image ; la deuxieme est une application de cette analyse a des images radiographiques d'os. L'afo d'une image consiste a estimer le parametre h du mouvement brownien fractionnaire (fbm) le long de lignes de niveaux de gris paralleles dans des directions differentes. Le but est double: chiffrer la rugosite de la texture et rendre compte de son anisotropie. Pour valider cette methode sur des images synthetiques, nous proposons tout d'abord une methode pour evaluer la qualite des techniques de synthese 1d du fbm. Puis, sur des signaux de reference, nous comparons differents estimateurs de h pour ne retenir que les meilleurs en termes de biais et de variance. Les methodes de synthese selectionnees dans le cas 1d sont ensuite etendues pour obtenir des textures isotropes et anisotropes. Ces images sont ensuite analysees par afo et les resultats sont reportes sur un diagramme polaire dont l'analyse harmonique permet de valider la methode et d'extraire des parametres caracteristiques de la texture. L'afo est ensuite utilisee pour caracteriser la texture des images osseuses et quantifier le processus de decalcification de l'os. En premier lieu, les images d'os de deux populations (temoins, osteoporotiques) sont analysees. Ensuite, des resultats issus de l'afo sont interpretes en liaison avec ceux d'une analyse histomorphometrique et ceux d'une evaluation de la resistance osseuse par des tests biomecaniques. Finalement, d'autres attributs de texture sont estimes sur nos images d'os par matrice de co-occurence et matrice de longueurs de plages. Quoique perfectible, l'outil que nous proposons semble prometteur en vue de quantifier les changements architecturaux de l'os trabeculaire lies a l'osteoporose
HAL (Le Centre pour la Communication Scientifique Directe), 1996
International audienceThe resistance of bone tissue is influenced not only by bone density parame... more International audienceThe resistance of bone tissue is influenced not only by bone density parameters but also by bone architecture parameters, such as the microarchitecture and anisotropy of trabecular bone. We have developed and validated a fractal analysis method for studying bone microarchitecture on roentgenograms. This technique provides reproducible measurements of the fractal dimension (D) of bone, which reflects bone texture. The fractal dimension is determined in 36 different directions; the mean of these 36 values is representative of the image. A polar diagram gives the value of D according to the angle of analysis. By decomposing this diagram using polar Fourier Transform analysis, the parameters related to the shape of the polar diagram can be determined. This diagram image analysis technique has been used for other similar diagrams and applied to the results of our fractal analysis method. Diagram shape characterization may provide information on the angular distribution of results and therefore on the anisotropy of the images under study. The purpose of this study was to compare roentgenograms of the calcaneus and radius in the same subjects to determine whether texture and anisotropy parameters discriminated between these two bones. Roentgenograms of the calcaneus and radius were obtained in ten nonosteoporotic subjects. The radius had a smaller fractal dimension than the calcaneus (mean +/- standard deviation: 1.215 +/- 0.025 and 1.285 +/- 0.066, respectively; p = 0.014). Differences in the shape of the polar diagram were found between the two bones. The mean Fourier coefficient ratio C2/C4 was considerably smaller at the calcaneus (0.63 +/- 0.50) than at the radius (4.88 +/- 3.45; p = 0.005). Our method allows quantitative characterization of texture and anisotropy differences between the calcaneus and radius. The smaller fractal dimension of the radius probably reflects the simpler architecture of this non weight-bearing bone. The differences in polar diagram shape allow to evaluate anisotropy differences between the calcaneus and radius
HAL (Le Centre pour la Communication Scientifique Directe), 1999
International audienceThe overall goal of this project is to propose a non-invasive method for a ... more International audienceThe overall goal of this project is to propose a non-invasive method for a precise evaluation of bone architectural changes during osteoporosis. Our approach is based on a fractal analysis of the anisotropic texture of trabecular bone radiographs. Texture roughness is characterised by the fractional Brownian motion (FBM) model of parameter H. First, eight estimation methods for H were compared using reference FBM signals. It is shown that the maximum likelihood estimator (MLE) method provides the most accurate estimate on which to base a fractal analysis. Two parameters were derived: the first one quantified the average roughness of the texture while the other analysed roughness relative to its anisotropy. Subsequently, the image acquisition process used was characterised and optimised in order to ensure a satisfactory long-term reproducibility. Finally, the fractal analysis using MLE was applied to clinical data. It was found that the mean roughness was significantly different on osteoporotic and control cases. It was also demonstrated that mean roughness was also related to histomorphometric measures of the two-dimensional bone architecture. Finally, our research demonstrated that anisotropy parameter and bone strength parameters recovered after compression tests were correlated
2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), Mar 28, 2022
Knee OsteoArthritis (OA) is a common musculoskeletal disorder, which causes reduced mobility for ... more Knee OsteoArthritis (OA) is a common musculoskeletal disorder, which causes reduced mobility for seniors. Due to the semi-quantitative nature of the Kellgren-Lawrence (KL) grading system, medical practitioners’ grading is subjective, being entirely based on their experience. With the development of computer vision, Computer-Aided Diagnosis (CAD) systems based on deep learning methods such as convolutional neural networks (CNNs) have shown success in knee OA diagnosis. In this paper, we propose a new approach, the so-called Siamese-GAP Network, for the early detection of knee OA through a KL-grade classification. More precisely, a set of Global Average Pooling (GAP) layers is integrated into the Siamese network used to extract features from each level. The obtained features are then combined to improve the classification performance. Our experimental results on baseline X-ray images from the OsteoArthritis Initiative (OAI) dataset show that the proposed approach presents potential results for the early detection of knee OA.
Uploads
Papers by Rachid Jennane