The Experts below are selected from a list of 183 Experts worldwide ranked by ideXlab platform
Sebastien Ourselin - One of the best experts on this subject based on the ideXlab platform.
-
MIUA - An Adaptive Sampling Scheme to Efficiently Train Fully Convolutional Networks for Semantic Segmentation
Communications in Computer and Information Science, 2018Co-Authors: Lorenz Berger, Eoin R Hyde, M. Jorge Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: (1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. (2) We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. (3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
Boosted Training of Convolutional Neural Networks for Multi-Class Segmentation.
arXiv: Computer Vision and Pattern Recognition, 2018Co-Authors: Lorenz Berger, Eoin R Hyde, Matt Gibb, Nevil Pavithran, Garin Kelly, Faiz Mumtaz, Sebastien OurselinAbstract:Training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large volumetric data sets, such as CT scans. Our contribution is threefold: 1) We propose a boosted sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in a more informative loss. This results in a significant training speed up and improves learning performance for image segmentation. 2) We propose a novel algorithm for boosting the SGD learning rate schedule by adaptively increasing and lowering the learning rate, avoiding the need for extensive hyperparameter tuning. 3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
an adaptive sampling scheme to efficiently train fully convolutional networks for semantic segmentation
Annual Conference on Medical Image Understanding and Analysis, 2017Co-Authors: Lorenz Berger, Eoin R Hyde, Jorge M Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: (1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. (2) We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. (3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
A Self-aware Sampling Scheme to Efficiently Train Fully Convolutional Networks for Semantic Segmentation
arXiv: Computer Vision and Pattern Recognition, 2017Co-Authors: Lorenz Berger, Eoin R Hyde, M. Jorge Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: 1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
Van Wallach - One of the best experts on this subject based on the ideXlab platform.
-
a revision of the genus leptotyphlops in northeastern africa and southwestern arabia serpentes leptotyphlopidae
Zootaxa, 2007Co-Authors: Donald G Broadley, Van WallachAbstract:The species of Leptotyphlops in northeastern Africa and southwestern Arabia (i.e. between 18°N and 12°S and between 29°E and 52°E) are revised. Twenty-five species belonging to five species groups are recognised, the taxonomy being based largely on skull morphology and Visceral Anatomy. Seven new species ( L. ionidesi, L. tanae, L. howelli, L. mbanjensis, L. keniensis, L. aethiopicus and L. nigroterminus ) and one new subspecies (L. scutifrons pitmani ) are described. Wallach & Lanza (2004) have revived Leptotyphlops braccianii (Scortecci) from the synonymy of L. macrorhynchus (Jan & Sordelli), with Glauconia variabilis Scortecci regarded as a synonym. They also revived L. erythraeus (Scortecci) from the synonymy of L. macrorhynchus and L. yemenicus Scortecci from the synonymy of L. nursii. Leptotyphlops monticolus (Chabanaud) is now revived from the synonymy of L. emini (Boulenger). Glauconia latirostris Sternfeld is reinstated as a full species. Leptotyphlops nursii (Boulenger), long known from the Arabian peninsula, is recorded from higher altitudes in northern Somalia. Descriptions of the Visceral Anatomy are presented for all species found in eastern and southern Africa. A key based on external characters is provided, the known variation in meristic and Visceral characters is tabulated and the distributions mapped.
-
Reexamination of an Anomalous Distribution: Resurrection of Ramphotyphlops becki (Serpentes: Typhlopidae)
2000Co-Authors: Glenn Shea, Van WallachAbstract:Ramphotyphlopsbecki (Tanner, 1948), restricted to Guadalcanal, Solomon Islands, is resurrected from the synonymy of Ramphotyphlops willeyi (Boulenger, 1900), from the Loyalty Islands, on the basis of consistent differ ences in external morphology and Visceral Anatomy. New records of Ramphoty phlops braminus (Daudin, 1803) are reported from Vanuatu and the Loyalty Islands. THE TYPHLOPID SNAKES of the Solomon Is lands were revised by McDowell (1974), who recognized six species from the archipelago, all placed in a single genus (then Typhlina, now Ramphotyphlops (ICZN 1982)): R. affi nis (Boulenger, 1889), R angusticeps (peters, 1877), R. braminus (Daudin, 1803), R. fla viventer (Peters, 1864), R. subocularis (Waite, 1897), and R willeyi (Boulenger, 1900). McDowell's (1974) revision was valuable for its introduction of new characters into ty phlopid systematics and for the thorough lit erature review, but it suffered from the small samples available for many species. The largest sample from the archipelago was for R flaviventer (5); for two species, R brami nus and R. subocularis, the only examined specimens were extralimital to the Solomons. This paucity of material resulted in a conser vative bias to "... regard related forms as conspecific unless some compelling evidence indicates their distinctness...." (McDowell 1974: 3). Recent revisions of two of the Sol omons typhlopids using larger samples re vealed that this bias seriously underestimated typhlopid species diversity in the New Guinea-Solomons region (Wallach 1995, 1996), with McDowell's species being com posite. In this paper we consider the identity of a third Solomon Islands typhlopid, identi fied by McDowell as Typhlina willeyi.
-
Visceral Anatomy of the Malaysian snake genus Xenophidion, including a cladistic analysis and allocation to a new family (Serpentes: Xenophidiidae)
Amphibia-Reptilia, 1998Co-Authors: Van Wallach, Rainer GüntherAbstract:The internal Anatomy of Xenophidion is described and compared with that of members of other snake families. A suite of primitive characters eliminates Xenophidion as a possible member of the Caenophidia; only two characters could conceivably suggest a relationship to the Caenophidia and both are examples of losses and thus of low phylogenetic value in assessing relationships. However, among lower snakes a sister group relationship is demonstrated with the Tropidophiidae of the Neotropical region. Besides possessing nearly identical viscera and topographical arrangement thereof, Xenophidion shares several characters with the Tropidophiidae. A new family is created to contain the genus, the Xenophidiidae. The Xenophidiidae share one synapomorphy with both the Tropidophiidae and Bolyeriidae. Therefore, it is proposed that these three families be united in the superfamily Tropidophioidea. A phylogenetic analysis of 52 characters results in the following preferred hypothesis of relationships: (Boinae, (((Bolyeria, Casarea), (Xenophidion, ((Exiliboa, Ungaliophis), (Trachyboa, Tropidophis)))), Acrochordus)).
-
leptotyphlops broadleyi a new species of worm snake from cote d ivoire serpentes leptotyphlopidae
African Journal of Herpetology, 1997Co-Authors: Van Wallach, D. E. HahnAbstract:WALLACH, V., and D.E. HAHN. 1997. Leptotyphlops broadleyi, a new species of worm snake from Cote d'Ivoire (Serpentes: Leptotyphlopidae). Afr. J. Herpetol. 46(2): 103–109. A new species of diminutive West African Leptotyphlops is described, based upon a series of 13 specimens from south-central Cote d'Ivoire. It resembles L. bicolor but differs in having a lower number of mid-dorsal scales, a relatively longer tail with higher number of subcaudals, an enlarged, hexagonal prefrontal (broader than deep and at least twice as large as supraocular), a tail terminating in a vertically compressed ridge, a relatively stouter body, and a smaller average and maximum size. The Visceral Anatomy is described, based upon a male and female paratype, and a key to the Leptotyphlops of Cote d'Ivoire is presented.
João L. Vilaça - One of the best experts on this subject based on the ideXlab platform.
-
A novel multi-atlas strategy with dense deformation field reconstruction for abdominal and thoracic multi-organ segmentation from computed tomography.
Medical image analysis, 2018Co-Authors: Bruno Oliveira, Sandro Queirós, Pedro Morais, Helena R. Torres, João Gomes-fonseca, Jaime C. Fonseca, João L. VilaçaAbstract:Anatomical evaluation of multiple abdominal and thoracic organs is generally performed with computed tomography images. Owing to the large field-of-view of these images, automatic segmentation strategies are typically required, facilitating the clinical evaluation. Multi-atlas segmentation (MAS) strategies have been widely used with this process, requiring multiple alignments between the target image and the set of known datasets, and subsequently fusing the alignment results to obtain the final segmentation. Nonetheless, current MAS strategies apply a global alignment of a deformable object, per organ, subdividing the segmentation process into multiple ones and losing the spatial information among nearby organs. This paper presents a novel MAS approach. First, a coarse-to-fine method with multiple global alignments (one per organ) is used. To make the method spatially coherent, these individual organs' global transformations are then fused in one using a dense deformation field reconstruction strategy. Second, from the candidate segmentations obtained, the final segmentation is estimated through an organ-based label fusion approach. The proposed method is evaluated and compared against a conventional MAS strategy through the segmentation of twelve abdominal and thoracic organs from the Visceral Anatomy benchmark. Average Dice coefficients for the liver, spleen, lungs and kidneys are all higher than 90%, are around 85% for the aorta, trachea and sternum and 70% for the pancreas, urinary bladder and gallbladder. The novel MAS strategy, with dense deformation field reconstruction, shows competitive results against other state-of-the-art methods, proving its added value for the segmentation of abdominal and thoracic organs, mainly for highly variable organs.
Vilaça João - One of the best experts on this subject based on the ideXlab platform.
-
A novel multi-atlas strategy with dense deformation field reconstruction for abdominal and thoracic multi-organ segmentation from computed tomography
Medical Image Analysis, 2018Co-Authors: Oliveira Bruno, Queirós Sandro, Morais Pedro, Torres Helena, Fonseca João, Fonseca Jaime, Vilaça JoãoAbstract:Anatomical evaluation of multiple abdominal and thoracic organs is generally performed with computed tomography images. Owing to the large field-of-view of these images, automatic segmentation strategies are typically required, facilitating the clinical evaluation. Multi-atlas segmentation (MAS) strategies have been widely used with this process, requiring multiple alignments between the target image and the set of known datasets, and subsequently fusing the alignment results to obtain the final segmentation. Nonetheless, current MAS strategies apply a global alignment of a deformable object, per organ, subdividing the segmentation process into multiple ones and losing the spatial information among nearby organs. This paper presents a novel MAS approach. First, a coarse-to-fine method with multiple global alignments (one per organ) is used. To make the method spatially coherent, these individual organs’ global transformations are then fused in one using a dense deformation field reconstruction strategy. Second, from the candidate segmentations obtained, the final segmentation is estimated through an organ-based label fusion approach. The proposed method is evaluated and compared against a conventional MAS strategy through the segmentation of twelve abdominal and thoracic organs from the Visceral Anatomy bench- mark. Average Dice coefficients for the liver, spleen, lungs and kidneys are all higher than 90%, are around 85% for the aorta, trachea and sternum and 70% for the pancreas, urinary bladder and gallbladder. The novel MAS strategy, with dense deformation field reconstruction, shows competitive results against other state-of-the-art methods, proving its added value for the segmentation of abdominal and thoracic organs, mainly for highly variable organs.The authors acknowledge Fundação para a Ciência e a Tecnologia (FCT), Portugal and the European Social Found, European Union, for funding support through the “Programa Op- eracional Capital Humano” (POCH) in the scope of the PhD grants SFRH/BD/93443/2013 (S. Queirós), SFRH/BD/95438/2013 (P. Morais), and PD/BDE/113597/2015 (J. Gomes-Fonseca). Moreover, authors gratefully acknowledge the funding of the projects NORTE-01-0145-FEDER-000013 and NORTE-01-0145- FEDER-024300, supported by Northern Portugal Regional Operational Programme (NORTE 2020), under the Portugal 2020 Partnership Agreement, through the European Regional Development Fund (FEDER). This work has been funded by FEDER funds, through the Competitiveness Factors Operational Programme (COMPETE), and by National funds, through the Foundation for Science and Technology (FCT), under the scope of the project POCI-01-0145-FEDER- 007038
Lorenz Berger - One of the best experts on this subject based on the ideXlab platform.
-
MIUA - An Adaptive Sampling Scheme to Efficiently Train Fully Convolutional Networks for Semantic Segmentation
Communications in Computer and Information Science, 2018Co-Authors: Lorenz Berger, Eoin R Hyde, M. Jorge Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: (1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. (2) We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. (3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
Boosted Training of Convolutional Neural Networks for Multi-Class Segmentation.
arXiv: Computer Vision and Pattern Recognition, 2018Co-Authors: Lorenz Berger, Eoin R Hyde, Matt Gibb, Nevil Pavithran, Garin Kelly, Faiz Mumtaz, Sebastien OurselinAbstract:Training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large volumetric data sets, such as CT scans. Our contribution is threefold: 1) We propose a boosted sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in a more informative loss. This results in a significant training speed up and improves learning performance for image segmentation. 2) We propose a novel algorithm for boosting the SGD learning rate schedule by adaptively increasing and lowering the learning rate, avoiding the need for extensive hyperparameter tuning. 3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
an adaptive sampling scheme to efficiently train fully convolutional networks for semantic segmentation
Annual Conference on Medical Image Understanding and Analysis, 2017Co-Authors: Lorenz Berger, Eoin R Hyde, Jorge M Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: (1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. (2) We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. (3) We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
-
A Self-aware Sampling Scheme to Efficiently Train Fully Convolutional Networks for Semantic Segmentation
arXiv: Computer Vision and Pattern Recognition, 2017Co-Authors: Lorenz Berger, Eoin R Hyde, M. Jorge Cardoso, Sebastien OurselinAbstract:Deep convolutional neural networks (CNNs) have shown excellent performance in object recognition tasks and dense classification problems such as semantic segmentation. However, training deep neural networks on large and sparse datasets is still challenging and can require large amounts of computation and memory. In this work, we address the task of performing semantic segmentation on large data sets, such as three-dimensional medical images. We propose an adaptive sampling scheme that uses a-posterior error maps, generated throughout training, to focus sampling on difficult regions, resulting in improved learning. Our contribution is threefold: 1) We give a detailed description of the proposed sampling algorithm to speed up and improve learning performance on large images. We propose a deep dual path CNN that captures information at fine and coarse scales, resulting in a network with a large field of view and high resolution outputs. We show that our method is able to attain new state-of-the-art results on the Visceral Anatomy benchmark.
