Joselene Marques | University of Copenhagen (original) (raw)

Papers by Joselene Marques

Research paper thumbnail of 422 Automatic Segmentation of Bone and Cartilage from Knee Mri

Osteoarthritis and Cartilage

ABSTRACT

Research paper thumbnail of 149 Diagnosis of Osteoarthrhis by Quantification of Tibia Trabecular Bone Structure from Magnetic Resonance Imaging

Osteoarthritis and Cartilage

Poster Presentations / Osteoarthritis and Cartilage 19S1 (2011) S53-S236 S75 of CCL3 and CCL4 wer... more Poster Presentations / Osteoarthritis and Cartilage 19S1 (2011) S53-S236 S75 of CCL3 and CCL4 were over 21 ng/L in normal adults, which were much higher than IL-6 and IL-1b ( ). The concentration of CCL3 was 36.34±2.68 ng/l, 55.30±4.05 ng/l, and 62.56±2.39 ng/l in normal adults, medial OA patients, and severe OA patients ( ). The ratio of CCL4 was significant different in normal adults and different stages of OA patients , and the concentration was 21.75±4.80 ng/l, 43.80±4.27 ng/l, and 60.50±5.87 ng/l ( ). Conclusion: CCL3, CCL4, CXCL2, IL-8 and IL-1b could be serum biomarkers to identify healthy adults, early stage OA patients and late stage OA patients from our primary data, especially CCL4 and CXCL2 were detectable and useful.

Research paper thumbnail of 7 Prognosis of Cartilage Loss by Analyzing Magnetic Resonance Imaging of the Tibia Trabecular Bone Structure

Osteoarthritis and Cartilage

such as depression, can increase an individual's experience of pain. We used factor analysis to d... more such as depression, can increase an individual's experience of pain. We used factor analysis to determine the relationships among symptomatic and psychosocial variables in a community-based sample with and without OA. Methods: We used data from individuals who were both enrolled in both the Johnston County OA Project and the separately funded Arthritis, Coping, & Emotion (ACE) study (data collected 1999-2005, n = 2239). This sample was approximately 1/3 male and African American, with mean age of 65±11 years, body mass index 30±7 kg/m2, and 13±11 years of education. Half had radiographic OA (defined as a Kellgren-Lawrence grade ≥2) of at least one hip or knee. Symptomatic data were available for 9 joint sites: the lower back and bilateral hands, knees, hips, and feet, and were graded as none, mild, moderate, or severe "on most days." The 7 validated psychosocial scales used were: the Centers for Epidemiological Studies Depression Scale (CES-D), Positive and Negative Affect Scale (PANAS), Rheumatology Attitudes Index (RAI), Social Support, Pain Catastrophizing Helplessness Subscale, Arthritis Impact Measurement Scale 2 Tension and Anxiety Subscale (AIMS2), and Life Orientation Test (LOT). Symptomatic data and the 7 scales were individually factor analyzed and appropriate composite scores were constructed. Higher order factor analysis was used to determine the relationships among these 8 composite scores. Cronbach's alpha was calculated as a measure of reliability and internal consistency. Analyses stratified by gender, race (African American vs. Caucasian), age (<55, 55-65, 65-75, 75+ years), body mass index (BMI, <25, 25 to <30, 30+ kg/meter squared), education (less than, equal to, or more than 12 years) and by radiographic OA status in the hip or knee were performed to determine any subgroup differences. Results: Analysis of each of the individual scales resulted in a single factor (all alphas >0.79) with the exception of the PANAS, which, as expected, had 2 factors (alpha 0.90) reflecting positive and negative affect characteristics. Symptoms from 9 joint sites also loaded onto a single factor (alpha 0.86). Higher order factor analysis using composite scores for each of these factors produced a single factor with an eigenvalue of 3.73 (See figure, due to missing values in individual scales, n = 1332, loadings 0.40-0.88, alpha 0.84). Dropping each score individually did not substantially change the alpha. Consistent results (factor loading pattern and alpha) were obtained when stratified by gender, race, age, BMI, education, or OA status. Figure: Screeplot of higher order factor and table of loadings of each scale onto this factor. *The eigenvalue represents the amount of information contained in a factor, while loadings represent the association between the individual scores and the factor.

Research paper thumbnail of Fighting the Semantic Gap on CBIR Systems through New Relevance Feedback Techniques

Proceedings of the Ieee Symposium on Computer Based Medical Systems, 2006

This paper introduces two novel relevance feedback techniques that integrate a new way to impleme... more This paper introduces two novel relevance feedback techniques that integrate a new way to implement the query center movement with a suitable weighting on the similarity function. These techniques integrated to a content-based image retrieval (CBIR) system, improves the precision of ...

Research paper thumbnail of Diagnosis of osteoarthritis and prognosis of tibial cartilage loss by quantification of tibia trabecular bone from MRI

Magnetic Resonance in Medicine, 2013

A longitudinal study was used to investigate the quantification of osteoarthritis and prediction ... more A longitudinal study was used to investigate the quantification of osteoarthritis and prediction of tibial cartilage loss by analysis of the tibia trabecular bone from magnetic resonance images of knees. The Kellgren Lawrence (KL) grades were determined by radiologists and the levels of cartilage loss were assessed by a segmentation process. Aiming to quantify and potentially capture the structure of the trabecular bone anatomy, a machine learning approach used a set of texture features for training a classifier to recognize the trabecular bone of a knee with radiographic osteoarthritis. Using crossvalidation, the bone structure marker was used to estimate for each knee both the probability of having radiographic osteoarthritis (KL >1) and the probability of rapid cartilage volume loss. The diagnostic ability reached a median area under the receiver-operator-characteristics curve of 0.92 (P < 0.0001), and the prognosis had odds ratio of 3.9 (95% confidence interval: 2.4-6.5). The medians of cartilage loss of the subjects classified as slow and rapid progressors were 1.1% and 4.9% per year, respectively. A preliminary radiological reading of the high and low risk knees put forward an hypothesis of which pathologies the bone marker could be capturing to define the prognosis of cartilage loss. Magn Reson Med 000:000-000,

Research paper thumbnail of Linear feature selection in texture analysis - A PLS based method

Machine Vision and Applications, 2013

ABSTRACT

Research paper thumbnail of A Percepção Humana na Recuperação de Imagens por Conteúdo

The Relevance Feedback (RR) approach is a powerful mechanism to refine and improve the techniques... more The Relevance Feedback (RR) approach is a powerful mechanism to refine and improve the techniques for content-based image retrieval (CBIR) considering the subjectivity introduced by the human analysis. Traditionally, in this process the human analyst weighs the images retrieved, considering their degree of relevance to the query posed. By doing so, the subjectivity of human perception is introduced in the CBIR, and the semantic gap inherent to this process is diminished. This work discusses the use of Relevance Feedback in a real ...

Research paper thumbnail of Fully automatic cartilage morphometry for knee MRI from the OAI

Research paper thumbnail of Automatic segmentation of bone and cartilage from knee MRI

Research paper thumbnail of Diagnosis of osteoarthritis by quantification of Tibia trabecular bone structure from magnetic resonance imaging

Research paper thumbnail of Prediction of cartilage loss by analysis of trabecular Tibia bone structure

Research paper thumbnail of Quantification of trabecular Tibia bone structure related to the presence of Osteoarthritis

Research paper thumbnail of Automatic segmentation of high- and low-field knee MRIs using knee image quantification with data from the osteoarthritis initiative

Journal of Medical Imaging, 2015

Research paper thumbnail of Um Novo Método Para Maximizar a Precisão Das Buscas Por Conteúdo Em Imagens Médicas

ime.uerj.br

... por Conteúdo em Imagens Médicas Marcela X. Ribeiro, Joselene Marques, Agma JM Traina, Caetano... more ... por Conteúdo em Imagens Médicas Marcela X. Ribeiro, Joselene Marques, Agma JM Traina, Caetano Traina Jr Instituto de Ciências Matemáticas e de Computação Universidade de São Paulo (USP) – São Carlos, SP – Brasil {mxavier,joselene,agma,caetano}@icmc.usp.br ...

Research paper thumbnail of Combatendo os Pesadelos da Busca Por Conteúdo em Imagens Médicas

Resumo - Os dois principais pesadelos que diminuem a qualidade da busca por conteúdo são: a) a &q... more Resumo - Os dois principais pesadelos que diminuem a qualidade da busca por conteúdo são: a) a "maldição da alta dimensionalidade", que degrada as estruturas de índice e diminui o poder de discriminação das características extraídas das imagens e b) o "gap semântico" existente entre a representação das características de baixo nível e sua interpretação humana. Neste artigo é proposto um novo método para aumentar a precisão das buscas por conteúdo de imagens médicas que combina técnicas de mineração de regras de associação e de realimentação de relevância. Regras de associação estatísticas são usadas para selecionar as características com maior poder de discriminação das imagens lidando com o problema da maldição da alta dimensionalidade. Uma técnica eficiente de realimentação de relevância é usada para lidar com o problema do gap semântico. Experimentos mostram que o método proposto é eficaz levando a um aumento na precisão das buscas de até 100%. Palavras-chave:...

Research paper thumbnail of Statistical Association Rules and Relevance Feedback: Powerful Allies to Improve the Retrieval of Medical Images

19th IEEE Symposium on Computer-Based Medical Systems (CBMS'06), 2006

Abstract This work aims at developing an efficient support to improve the precision of medical im... more Abstract This work aims at developing an efficient support to improve the precision of medical image retrieval by content, introducing an approach that combines techniques of statistical association rule mining and relevance feedback. Low level features of shape and texture are ...

Research paper thumbnail of Osteoarthritis Imaging by Quantification of Tibial Trabecular Bone

The pathogenesis of osteoarthritis (OA) includes complex events in the whole joint. Cartilage los... more The pathogenesis of osteoarthritis (OA) includes complex events in the whole joint. Cartilage loss and bone remodelling are central in OA progression. In this project, we investigated the feasibility of quantifying OA by analysis of the tibial trabecular bone structure in low-field knee magnetic resonance imaging (MRI). The development of automatic and more sensitive indicators of OA in conjunction with low cost equipment have the potential to decrease the length and cost of clinical trials. We present a texture analysis methodology that combined uncommitted machine-learning techniques in a fully automatic framework. Different linear feature selection approaches where investigated. The methodology was evaluated in a longitudinal study, where MRI scans of knees were used to quantify the tibial trabecular bone in a bone marker for OA diagnosis and another marker for prediction of tibial cartilage loss. The healthy and diseased subjects were defined by the Kellgren and Lawrence index assigned by radiologists and the levels of cartilage loss were assessed by a segmentation process. A preliminary radiological reading of the knees with high and low risks of cartilage loss suggested the prognosis bone marker captured aspects of the vertical trabecularization of the tibial bone to define the prognosis of cartilage loss. We also investigated which region of the tibia provided the best prognosis for medial tibial cartilage loss. The structure of the tibial trabecular bone was divided in localized subregions in an attempt to capture the different pathological features occurring at each location. We applied multiple-instance learning, where each subregion was defined to be one instance and a bag held all instances over a full region-of-interest. The inferior part of the tibial bone was classified as the most relevant region for prognosis of cartilage loss and a preliminary radiological reading of a subset of the samples suggested the bone marker also captured the vertical trabecularization of the tibial bone to define the most relevant region. In a clinical point of view, besides presenting a bone marker able to predict disease progression and diagnostic bone marker superior to other OA biomarkers, our findings underlined the importance of the trabecular bone to the understanding of the OA pathology.

Research paper thumbnail of Diagnosis and prognosis of osteoarthritis by texture analysis using sparse linear models

Abstract. We present a texture analysis methodology that combines uncommitted machine-learning te... more Abstract. We present a texture analysis methodology that combines uncommitted machine-learning techniques and sparse feature transformation methods in a fully automatic framework. We compare the performances of a partial least squares (PLS) forward feature selection strategy to a hard threshold sparse PLS algorithm and a sparse linear discriminant model. The texture analysis framework was applied to diagnosis of knee osteoarthritis (OA) and prognosis of cartilage loss.

Research paper thumbnail of Linear feature selection in texture analysis-A PLS based method

Machine Vision and Applications, Jan 1, 2012

We present a texture analysis methodology that combined uncommitted machine-learning techniques a... more We present a texture analysis methodology that combined uncommitted machine-learning techniques and partial least square (PLS) in a fully automatic framework. Our approach introduces a robust PLS-based dimensionality reduction (DR) step to specifically address outliers and high-dimensional feature sets. The texture analysis framework was applied to diagnosis of knee osteoarthritis (OA). To classify between healthy subjects and OA patients, a generic bank of texture features was extracted from magnetic resonance images of tibial knee bone. The features were used as input to the DR algorithm, which first applied a PLS regression to rank the features and then defined the best number of features to retain in the model by an iterative learning phase. The outliers in the dataset, that could inflate the number of selected features, were eliminated by a pre-processing step. To cope with the limited number of samples, the data were evaluated using Monte Carlo cross validation (CV). The developed DR method demonstrated consistency in selecting a relatively homogeneous set of features across the CV iterations. Per each CV group, a median of 19 % of the original features was selected and considering all CV groups, the methods selected 36 % of the original features available. The diagnosis evaluation reached a generalization area-under-the-ROC curve of 0.92, which was higher than established cartilage-based markers known to relate to OA diagnosis.

Research paper thumbnail of Realimentação de Relevância: Integração do Conhecimento do especialista com a Recuperação de Imagens por Conteúdo

As técnicas de Realimentação de Relevância introduzem o usuário no processo de busca por similari... more As técnicas de Realimentação de Relevância introduzem o usuário no processo de busca por similaridade de imagens baseadas em conteúdo. A iteração do usuário com o sistema permite trazer o conhecimento do especialista para o processo de representação da imagem exemplo usada como centro da consulta. Neste artigo propomos duas novas técnicas de Realimentação de Relevância. Os experimentos mostraram que as técnicas melhoram as consultas em até 45% após 5 iterações. A análise dos dados coletados a partir de experimentos com usuários mostrou que o reprocessamento da consulta leva menos de 1 segundo e o grau de satisfação com os resultados foi de 80% após 3 iterações em média. This paper introduces two novel relevance feedback (RF) techniques that integrated to a content-based image retrieval system improves the precision of the results up to 45% employing 5 iterations. Besides being effective, the techniques are efficient as they take less than one second to reprocess the queries at each iteration. The experiments show that the number of feedback iterations should go to up 3, but the first one holds the major gain in improvement. The user satisfaction achieved 80% after an average of 3 RF cicles.

Research paper thumbnail of 422 Automatic Segmentation of Bone and Cartilage from Knee Mri

Osteoarthritis and Cartilage

ABSTRACT

Research paper thumbnail of 149 Diagnosis of Osteoarthrhis by Quantification of Tibia Trabecular Bone Structure from Magnetic Resonance Imaging

Osteoarthritis and Cartilage

Poster Presentations / Osteoarthritis and Cartilage 19S1 (2011) S53-S236 S75 of CCL3 and CCL4 wer... more Poster Presentations / Osteoarthritis and Cartilage 19S1 (2011) S53-S236 S75 of CCL3 and CCL4 were over 21 ng/L in normal adults, which were much higher than IL-6 and IL-1b ( ). The concentration of CCL3 was 36.34±2.68 ng/l, 55.30±4.05 ng/l, and 62.56±2.39 ng/l in normal adults, medial OA patients, and severe OA patients ( ). The ratio of CCL4 was significant different in normal adults and different stages of OA patients , and the concentration was 21.75±4.80 ng/l, 43.80±4.27 ng/l, and 60.50±5.87 ng/l ( ). Conclusion: CCL3, CCL4, CXCL2, IL-8 and IL-1b could be serum biomarkers to identify healthy adults, early stage OA patients and late stage OA patients from our primary data, especially CCL4 and CXCL2 were detectable and useful.

Research paper thumbnail of 7 Prognosis of Cartilage Loss by Analyzing Magnetic Resonance Imaging of the Tibia Trabecular Bone Structure

Osteoarthritis and Cartilage

such as depression, can increase an individual's experience of pain. We used factor analysis to d... more such as depression, can increase an individual's experience of pain. We used factor analysis to determine the relationships among symptomatic and psychosocial variables in a community-based sample with and without OA. Methods: We used data from individuals who were both enrolled in both the Johnston County OA Project and the separately funded Arthritis, Coping, & Emotion (ACE) study (data collected 1999-2005, n = 2239). This sample was approximately 1/3 male and African American, with mean age of 65±11 years, body mass index 30±7 kg/m2, and 13±11 years of education. Half had radiographic OA (defined as a Kellgren-Lawrence grade ≥2) of at least one hip or knee. Symptomatic data were available for 9 joint sites: the lower back and bilateral hands, knees, hips, and feet, and were graded as none, mild, moderate, or severe "on most days." The 7 validated psychosocial scales used were: the Centers for Epidemiological Studies Depression Scale (CES-D), Positive and Negative Affect Scale (PANAS), Rheumatology Attitudes Index (RAI), Social Support, Pain Catastrophizing Helplessness Subscale, Arthritis Impact Measurement Scale 2 Tension and Anxiety Subscale (AIMS2), and Life Orientation Test (LOT). Symptomatic data and the 7 scales were individually factor analyzed and appropriate composite scores were constructed. Higher order factor analysis was used to determine the relationships among these 8 composite scores. Cronbach's alpha was calculated as a measure of reliability and internal consistency. Analyses stratified by gender, race (African American vs. Caucasian), age (<55, 55-65, 65-75, 75+ years), body mass index (BMI, <25, 25 to <30, 30+ kg/meter squared), education (less than, equal to, or more than 12 years) and by radiographic OA status in the hip or knee were performed to determine any subgroup differences. Results: Analysis of each of the individual scales resulted in a single factor (all alphas >0.79) with the exception of the PANAS, which, as expected, had 2 factors (alpha 0.90) reflecting positive and negative affect characteristics. Symptoms from 9 joint sites also loaded onto a single factor (alpha 0.86). Higher order factor analysis using composite scores for each of these factors produced a single factor with an eigenvalue of 3.73 (See figure, due to missing values in individual scales, n = 1332, loadings 0.40-0.88, alpha 0.84). Dropping each score individually did not substantially change the alpha. Consistent results (factor loading pattern and alpha) were obtained when stratified by gender, race, age, BMI, education, or OA status. Figure: Screeplot of higher order factor and table of loadings of each scale onto this factor. *The eigenvalue represents the amount of information contained in a factor, while loadings represent the association between the individual scores and the factor.

Research paper thumbnail of Fighting the Semantic Gap on CBIR Systems through New Relevance Feedback Techniques

Proceedings of the Ieee Symposium on Computer Based Medical Systems, 2006

This paper introduces two novel relevance feedback techniques that integrate a new way to impleme... more This paper introduces two novel relevance feedback techniques that integrate a new way to implement the query center movement with a suitable weighting on the similarity function. These techniques integrated to a content-based image retrieval (CBIR) system, improves the precision of ...

Research paper thumbnail of Diagnosis of osteoarthritis and prognosis of tibial cartilage loss by quantification of tibia trabecular bone from MRI

Magnetic Resonance in Medicine, 2013

A longitudinal study was used to investigate the quantification of osteoarthritis and prediction ... more A longitudinal study was used to investigate the quantification of osteoarthritis and prediction of tibial cartilage loss by analysis of the tibia trabecular bone from magnetic resonance images of knees. The Kellgren Lawrence (KL) grades were determined by radiologists and the levels of cartilage loss were assessed by a segmentation process. Aiming to quantify and potentially capture the structure of the trabecular bone anatomy, a machine learning approach used a set of texture features for training a classifier to recognize the trabecular bone of a knee with radiographic osteoarthritis. Using crossvalidation, the bone structure marker was used to estimate for each knee both the probability of having radiographic osteoarthritis (KL >1) and the probability of rapid cartilage volume loss. The diagnostic ability reached a median area under the receiver-operator-characteristics curve of 0.92 (P < 0.0001), and the prognosis had odds ratio of 3.9 (95% confidence interval: 2.4-6.5). The medians of cartilage loss of the subjects classified as slow and rapid progressors were 1.1% and 4.9% per year, respectively. A preliminary radiological reading of the high and low risk knees put forward an hypothesis of which pathologies the bone marker could be capturing to define the prognosis of cartilage loss. Magn Reson Med 000:000-000,

Research paper thumbnail of Linear feature selection in texture analysis - A PLS based method

Machine Vision and Applications, 2013

ABSTRACT

Research paper thumbnail of A Percepção Humana na Recuperação de Imagens por Conteúdo

The Relevance Feedback (RR) approach is a powerful mechanism to refine and improve the techniques... more The Relevance Feedback (RR) approach is a powerful mechanism to refine and improve the techniques for content-based image retrieval (CBIR) considering the subjectivity introduced by the human analysis. Traditionally, in this process the human analyst weighs the images retrieved, considering their degree of relevance to the query posed. By doing so, the subjectivity of human perception is introduced in the CBIR, and the semantic gap inherent to this process is diminished. This work discusses the use of Relevance Feedback in a real ...

Research paper thumbnail of Fully automatic cartilage morphometry for knee MRI from the OAI

Research paper thumbnail of Automatic segmentation of bone and cartilage from knee MRI

Research paper thumbnail of Diagnosis of osteoarthritis by quantification of Tibia trabecular bone structure from magnetic resonance imaging

Research paper thumbnail of Prediction of cartilage loss by analysis of trabecular Tibia bone structure

Research paper thumbnail of Quantification of trabecular Tibia bone structure related to the presence of Osteoarthritis

Research paper thumbnail of Automatic segmentation of high- and low-field knee MRIs using knee image quantification with data from the osteoarthritis initiative

Journal of Medical Imaging, 2015

Research paper thumbnail of Um Novo Método Para Maximizar a Precisão Das Buscas Por Conteúdo Em Imagens Médicas

ime.uerj.br

... por Conteúdo em Imagens Médicas Marcela X. Ribeiro, Joselene Marques, Agma JM Traina, Caetano... more ... por Conteúdo em Imagens Médicas Marcela X. Ribeiro, Joselene Marques, Agma JM Traina, Caetano Traina Jr Instituto de Ciências Matemáticas e de Computação Universidade de São Paulo (USP) – São Carlos, SP – Brasil {mxavier,joselene,agma,caetano}@icmc.usp.br ...

Research paper thumbnail of Combatendo os Pesadelos da Busca Por Conteúdo em Imagens Médicas

Resumo - Os dois principais pesadelos que diminuem a qualidade da busca por conteúdo são: a) a &q... more Resumo - Os dois principais pesadelos que diminuem a qualidade da busca por conteúdo são: a) a "maldição da alta dimensionalidade", que degrada as estruturas de índice e diminui o poder de discriminação das características extraídas das imagens e b) o "gap semântico" existente entre a representação das características de baixo nível e sua interpretação humana. Neste artigo é proposto um novo método para aumentar a precisão das buscas por conteúdo de imagens médicas que combina técnicas de mineração de regras de associação e de realimentação de relevância. Regras de associação estatísticas são usadas para selecionar as características com maior poder de discriminação das imagens lidando com o problema da maldição da alta dimensionalidade. Uma técnica eficiente de realimentação de relevância é usada para lidar com o problema do gap semântico. Experimentos mostram que o método proposto é eficaz levando a um aumento na precisão das buscas de até 100%. Palavras-chave:...

Research paper thumbnail of Statistical Association Rules and Relevance Feedback: Powerful Allies to Improve the Retrieval of Medical Images

19th IEEE Symposium on Computer-Based Medical Systems (CBMS'06), 2006

Abstract This work aims at developing an efficient support to improve the precision of medical im... more Abstract This work aims at developing an efficient support to improve the precision of medical image retrieval by content, introducing an approach that combines techniques of statistical association rule mining and relevance feedback. Low level features of shape and texture are ...

Research paper thumbnail of Osteoarthritis Imaging by Quantification of Tibial Trabecular Bone

The pathogenesis of osteoarthritis (OA) includes complex events in the whole joint. Cartilage los... more The pathogenesis of osteoarthritis (OA) includes complex events in the whole joint. Cartilage loss and bone remodelling are central in OA progression. In this project, we investigated the feasibility of quantifying OA by analysis of the tibial trabecular bone structure in low-field knee magnetic resonance imaging (MRI). The development of automatic and more sensitive indicators of OA in conjunction with low cost equipment have the potential to decrease the length and cost of clinical trials. We present a texture analysis methodology that combined uncommitted machine-learning techniques in a fully automatic framework. Different linear feature selection approaches where investigated. The methodology was evaluated in a longitudinal study, where MRI scans of knees were used to quantify the tibial trabecular bone in a bone marker for OA diagnosis and another marker for prediction of tibial cartilage loss. The healthy and diseased subjects were defined by the Kellgren and Lawrence index assigned by radiologists and the levels of cartilage loss were assessed by a segmentation process. A preliminary radiological reading of the knees with high and low risks of cartilage loss suggested the prognosis bone marker captured aspects of the vertical trabecularization of the tibial bone to define the prognosis of cartilage loss. We also investigated which region of the tibia provided the best prognosis for medial tibial cartilage loss. The structure of the tibial trabecular bone was divided in localized subregions in an attempt to capture the different pathological features occurring at each location. We applied multiple-instance learning, where each subregion was defined to be one instance and a bag held all instances over a full region-of-interest. The inferior part of the tibial bone was classified as the most relevant region for prognosis of cartilage loss and a preliminary radiological reading of a subset of the samples suggested the bone marker also captured the vertical trabecularization of the tibial bone to define the most relevant region. In a clinical point of view, besides presenting a bone marker able to predict disease progression and diagnostic bone marker superior to other OA biomarkers, our findings underlined the importance of the trabecular bone to the understanding of the OA pathology.

Research paper thumbnail of Diagnosis and prognosis of osteoarthritis by texture analysis using sparse linear models

Abstract. We present a texture analysis methodology that combines uncommitted machine-learning te... more Abstract. We present a texture analysis methodology that combines uncommitted machine-learning techniques and sparse feature transformation methods in a fully automatic framework. We compare the performances of a partial least squares (PLS) forward feature selection strategy to a hard threshold sparse PLS algorithm and a sparse linear discriminant model. The texture analysis framework was applied to diagnosis of knee osteoarthritis (OA) and prognosis of cartilage loss.

Research paper thumbnail of Linear feature selection in texture analysis-A PLS based method

Machine Vision and Applications, Jan 1, 2012

We present a texture analysis methodology that combined uncommitted machine-learning techniques a... more We present a texture analysis methodology that combined uncommitted machine-learning techniques and partial least square (PLS) in a fully automatic framework. Our approach introduces a robust PLS-based dimensionality reduction (DR) step to specifically address outliers and high-dimensional feature sets. The texture analysis framework was applied to diagnosis of knee osteoarthritis (OA). To classify between healthy subjects and OA patients, a generic bank of texture features was extracted from magnetic resonance images of tibial knee bone. The features were used as input to the DR algorithm, which first applied a PLS regression to rank the features and then defined the best number of features to retain in the model by an iterative learning phase. The outliers in the dataset, that could inflate the number of selected features, were eliminated by a pre-processing step. To cope with the limited number of samples, the data were evaluated using Monte Carlo cross validation (CV). The developed DR method demonstrated consistency in selecting a relatively homogeneous set of features across the CV iterations. Per each CV group, a median of 19 % of the original features was selected and considering all CV groups, the methods selected 36 % of the original features available. The diagnosis evaluation reached a generalization area-under-the-ROC curve of 0.92, which was higher than established cartilage-based markers known to relate to OA diagnosis.

Research paper thumbnail of Realimentação de Relevância: Integração do Conhecimento do especialista com a Recuperação de Imagens por Conteúdo

As técnicas de Realimentação de Relevância introduzem o usuário no processo de busca por similari... more As técnicas de Realimentação de Relevância introduzem o usuário no processo de busca por similaridade de imagens baseadas em conteúdo. A iteração do usuário com o sistema permite trazer o conhecimento do especialista para o processo de representação da imagem exemplo usada como centro da consulta. Neste artigo propomos duas novas técnicas de Realimentação de Relevância. Os experimentos mostraram que as técnicas melhoram as consultas em até 45% após 5 iterações. A análise dos dados coletados a partir de experimentos com usuários mostrou que o reprocessamento da consulta leva menos de 1 segundo e o grau de satisfação com os resultados foi de 80% após 3 iterações em média. This paper introduces two novel relevance feedback (RF) techniques that integrated to a content-based image retrieval system improves the precision of the results up to 45% employing 5 iterations. Besides being effective, the techniques are efficient as they take less than one second to reprocess the queries at each iteration. The experiments show that the number of feedback iterations should go to up 3, but the first one holds the major gain in improvement. The user satisfaction achieved 80% after an average of 3 RF cicles.