The Experts below are selected from a list of 207 Experts worldwide ranked by ideXlab platform
Carole Lartizien - One of the best experts on this subject based on the ideXlab platform.
-
regularized siamese neural network for unsupervised outlier detection on Brain multiparametric magnetic resonance imaging application to epilepsy lesion screening
Medical Image Analysis, 2020Co-Authors: Zaruhi Alaverdyan, Julien Jung, Romain Bouet, Carole LartizienAbstract:In this study, we propose a novel Anomaly detection model targeting subtle Brain lesions in multiparametric MRI. To compensate for the lack of annotated data adequately sampling the heterogeneity of such pathologies, we cast this problem as an outlier detection problem and introduce a novel configuration of unsupervised deep siamese networks to learn normal Brain representations using a series of non-pathological Brain scans. The proposed siamese network, composed of stacked convolutional autoencoders as subnetworks is designed to map patches extracted from healthy control scans only and centered at the same spatial localization to 'close' representations with respect to the chosen metric in a latent space. It is based on a novel loss function combining a similarity term and a regularization term compensating for the lack of dissimilar pairs. These latent representations are then fed into oc-SVM models at voxel-level to produce Anomaly score maps. We evaluate the performance of our Brain Anomaly detection model to detect subtle epilepsy lesions in multiparametric (T1-weighted, FLAIR) MRI exams considered as normal (MRI-negative). Our detection model trained on 75 healthy subjects and validated on 21 epilepsy patients (with 18 MRI-negatives) achieves a maximum sensitivity of 61% on the MRI-negative lesions, identified among the 5 most suspicious detections on average. It is shown to outperform detection models based on the same architecture but with stacked convolutional or Wasserstein autoencoders as unsupervised feature extraction mechanisms.
Zaruhi Alaverdyan - One of the best experts on this subject based on the ideXlab platform.
-
regularized siamese neural network for unsupervised outlier detection on Brain multiparametric magnetic resonance imaging application to epilepsy lesion screening
Medical Image Analysis, 2020Co-Authors: Zaruhi Alaverdyan, Julien Jung, Romain Bouet, Carole LartizienAbstract:In this study, we propose a novel Anomaly detection model targeting subtle Brain lesions in multiparametric MRI. To compensate for the lack of annotated data adequately sampling the heterogeneity of such pathologies, we cast this problem as an outlier detection problem and introduce a novel configuration of unsupervised deep siamese networks to learn normal Brain representations using a series of non-pathological Brain scans. The proposed siamese network, composed of stacked convolutional autoencoders as subnetworks is designed to map patches extracted from healthy control scans only and centered at the same spatial localization to 'close' representations with respect to the chosen metric in a latent space. It is based on a novel loss function combining a similarity term and a regularization term compensating for the lack of dissimilar pairs. These latent representations are then fed into oc-SVM models at voxel-level to produce Anomaly score maps. We evaluate the performance of our Brain Anomaly detection model to detect subtle epilepsy lesions in multiparametric (T1-weighted, FLAIR) MRI exams considered as normal (MRI-negative). Our detection model trained on 75 healthy subjects and validated on 21 epilepsy patients (with 18 MRI-negatives) achieves a maximum sensitivity of 61% on the MRI-negative lesions, identified among the 5 most suspicious detections on average. It is shown to outperform detection models based on the same architecture but with stacked convolutional or Wasserstein autoencoders as unsupervised feature extraction mechanisms.
Fabio Triulzi - One of the best experts on this subject based on the ideXlab platform.
-
response to is fetal magnetic resonance imaging indicated when ultrasound isolated mild ventriculomegaly is present in pregnancies with no risk factors
Prenatal Diagnosis, 2014Co-Authors: Cecilia Parazzini, Andrea Righini, Chiara Doneda, Mariangela Rustico, M Lanna, Fabio TriulziAbstract:Objective Ventriculomegaly (VM) is the most common Brain Anomaly in prenatal ultrasound (US) diagnosis. There is a general trend to perform fetal magnetic resonance imaging (MRI) when VM is severe (greater than 15 mm) and/or it is not isolated. The role of MRI is debated when VM is borderline (between 10 and 15 mm) and isolated. Some authors have subdivided borderline VM into mild (10 to 12 mm) and moderate (>12 to15 mm). The aim of the study was to evaluate the role of MR in the imaging protocol of fetal cases characterized by mild isolated VM and no risk factors. Method As a retrospective study, 179 fetal MRI exams (mean gestational age: 26 weeks), performed for mild, isolated VM on US, were analyzed to search additional or different findings with respect to ultrasound. The potential impact of MRI results on prenatal counselling is described. Results In 49/179 cases, MRI and US results differed, but only in two of these cases did MRI studies provide clinically consistent additional information. In 130/179 cases, MRI confirmed US findings. Conclusion In this extremely selected group of fetuses with isolated, mild VM and no risk factors, MRI may not be indicated in the prenatal imaging protocol. © 2012 John Wiley & Sons, Ltd.
-
Evidence of a congenital midline Brain Anomaly in pituitary dwarfs: a magnetic resonance imaging study in 101 patients.
Pediatrics, 1994Co-Authors: Fabio Triulzi, Giuseppe Scotti, Di Natale B, C. Pellini, Lukezic M, Scognamiglio M, Giuseppe ChiumelloAbstract:BACKGROUND: Magnetic resonance imaging (MRI) of the Brain in pituitary dwarfs has revealed a previously unknown entity: ectopia of the posterior pituitary (PPE), absence or hypoplasia of the pituitary stalk and hypoplasia of the anterior pituitary. The pathogenesis of these findings was explained originally by a traumatic transection of the pituitary stalk during delivery. A high incidence of breech delivery has been reported in these groups, but the traumatic hypothesis cannot explain the findings in the relatively high percentage of patients with normal delivery, nor account for a different feature also found in other pituitary dwarfs consisting of pituitary hypoplasia with normal posterior pituitary. A second hypothesis could then been proposed, based on dysgenesis or abnormal embryonic development of both adenohypophysis and neurohypophysis. OBJECTIVE: To review the value and significance of these two different etiopathogenetic hypotheses by analyzing clinical, endocrinological, and MRI findings in a large population of pituitary dwarfs. METHODS: One hundred and one consecutive patients with congenital idiopathic growth hormone deficiency (CIGHD) were studied by MRI; they were compared with a control group of 46 healthy short children. A complete clinico-endocrinological evaluation was obtained in both patients and controls to assess the perinatal history, the pituitary-hypothalamic function, and the neurological status. MRI studies were evaluated both qualitatively and quantitatively and the pituitary volume (PV) was calculated in both patients and controls. Quantitative data were statistically analyzed to compare the mean PV of the patients with the mean PV of controls, the hormonal therapy, the single or multiple pituitary hormone deficiency, and the presence of breech delivery. RESULTS: MRI revealed PPE in 59 patients and a normal posterior pituitary (NPP) in 42. PV was extremely small in patients with PPE and in patients with NPP associated with a severely narrowed pituitary stalk; mean PV was significantly lower in CIGHD patients when compared with that of healthy short children. PV was not influenced by hormonal therapy and did not differ between patients with single and multiple pituitary hormone deficiency and between patients with normal and breech delivery. PPE patients differed from NPP patients for a higher male/female ratio (3:1 vs 1:1) and for a greater frequency of multiple pituitary hormone deficiency (49% vs 12%), breech delivery (32% vs 7%), and associated congenital Brain anomalies (12% vs 7%). In PPE patients breech delivery was strongly associated with multiple pituitary hormone deficiency. CONCLUSION: On the basis of this study the traumatic hypothesis could theoretically explain the pathogenesis of PPE only in 32% of the patients with this condition. On the basis of modern understanding of embryogenesis of anterior and posterior pituitary, it is then justified to propose that a defective induction of mediobasal structure of the Brain in the early embryo could account for both the complex morphological MRI abnormality and the clinico-endocrinological features encountered in all PPE patients. The close contiguity between the future pituitary and hypothalamus, the peculiar association with congenital midline Brain anomalies, and the recent data about a possible role of Pit-1 gene, all support the hypothesis of a congenital defect. Finally, breech delivery can be considered not as a cause of PPE, but as an effect of the embryonic pituitary-hypothalamic abnormalities.
Julien Jung - One of the best experts on this subject based on the ideXlab platform.
-
regularized siamese neural network for unsupervised outlier detection on Brain multiparametric magnetic resonance imaging application to epilepsy lesion screening
Medical Image Analysis, 2020Co-Authors: Zaruhi Alaverdyan, Julien Jung, Romain Bouet, Carole LartizienAbstract:In this study, we propose a novel Anomaly detection model targeting subtle Brain lesions in multiparametric MRI. To compensate for the lack of annotated data adequately sampling the heterogeneity of such pathologies, we cast this problem as an outlier detection problem and introduce a novel configuration of unsupervised deep siamese networks to learn normal Brain representations using a series of non-pathological Brain scans. The proposed siamese network, composed of stacked convolutional autoencoders as subnetworks is designed to map patches extracted from healthy control scans only and centered at the same spatial localization to 'close' representations with respect to the chosen metric in a latent space. It is based on a novel loss function combining a similarity term and a regularization term compensating for the lack of dissimilar pairs. These latent representations are then fed into oc-SVM models at voxel-level to produce Anomaly score maps. We evaluate the performance of our Brain Anomaly detection model to detect subtle epilepsy lesions in multiparametric (T1-weighted, FLAIR) MRI exams considered as normal (MRI-negative). Our detection model trained on 75 healthy subjects and validated on 21 epilepsy patients (with 18 MRI-negatives) achieves a maximum sensitivity of 61% on the MRI-negative lesions, identified among the 5 most suspicious detections on average. It is shown to outperform detection models based on the same architecture but with stacked convolutional or Wasserstein autoencoders as unsupervised feature extraction mechanisms.
Romain Bouet - One of the best experts on this subject based on the ideXlab platform.
-
regularized siamese neural network for unsupervised outlier detection on Brain multiparametric magnetic resonance imaging application to epilepsy lesion screening
Medical Image Analysis, 2020Co-Authors: Zaruhi Alaverdyan, Julien Jung, Romain Bouet, Carole LartizienAbstract:In this study, we propose a novel Anomaly detection model targeting subtle Brain lesions in multiparametric MRI. To compensate for the lack of annotated data adequately sampling the heterogeneity of such pathologies, we cast this problem as an outlier detection problem and introduce a novel configuration of unsupervised deep siamese networks to learn normal Brain representations using a series of non-pathological Brain scans. The proposed siamese network, composed of stacked convolutional autoencoders as subnetworks is designed to map patches extracted from healthy control scans only and centered at the same spatial localization to 'close' representations with respect to the chosen metric in a latent space. It is based on a novel loss function combining a similarity term and a regularization term compensating for the lack of dissimilar pairs. These latent representations are then fed into oc-SVM models at voxel-level to produce Anomaly score maps. We evaluate the performance of our Brain Anomaly detection model to detect subtle epilepsy lesions in multiparametric (T1-weighted, FLAIR) MRI exams considered as normal (MRI-negative). Our detection model trained on 75 healthy subjects and validated on 21 epilepsy patients (with 18 MRI-negatives) achieves a maximum sensitivity of 61% on the MRI-negative lesions, identified among the 5 most suspicious detections on average. It is shown to outperform detection models based on the same architecture but with stacked convolutional or Wasserstein autoencoders as unsupervised feature extraction mechanisms.