Monte Carlo simulation is an essential tool in emission tomography that can assist in the design of new medical imaging devices, the optimization of acquisition protocols and the development or assessment of image reconstruction algorithms and correction techniques. GATE, the Geant4 Application for Tomographic Emission, encapsulates the Geant4 libraries to achieve a modular, versatile, scripted simulation toolkit adapted to the field of nuclear medicine. In particular, GATE allows the description of time-dependent phenomena such as source or detector movement, and source decay kinetics. This feature makes it possible to simulate time curves under realistic acquisition conditions and to test dynamic reconstruction algorithms. This paper gives a detailed description of the design and development of GATE by the OpenGATE collaboration, whose continuing objective is to improve, document and validate GATE by simulating commercially available imaging systems for PET and SPECT. Large effort is also invested in the ability and the flexibility to model novel detection systems or systems still under design. A public release of GATE licensed under the GNU Lesser General Public License can be downloaded at http:/www-lphe.epfl.ch/GATE/. Two benchmarks developed for PET and SPECT to test the installation of GATE and to serve as a tutorial for the users are presented. Extensive validation of the GATE simulation platform has been started, comparing simulations and measurements on commercially available acquisition systems. References to those results are listed. The future prospects towards the gridification of GATE and its extension to other domains such as dosimetry are also discussed.

No abstract

Due to the short and noisy nature of Twitter microposts, detecting named entities is often a cumbersome task. As part of the ACL2015 Named Entity Recognition (NER) shared task, we present a semisupervised system that detects 10 types of named entities. To that end, we leverage 400 million Twitter microposts to generate powerful word embeddings as input features and use a neural network to execute the classification. To further boost the performance, we employ dropout to train the network and leaky Rectified Linear Units (ReLUs). Our system achieved the fourth position in the final ranking, without using any kind of hand-crafted features such as lexical features or gazetteers.

No abstract