Search

Artificial intelligence is paving the way for a new era of algorithms focusing directly on the information contained in the data, autonomously extracting relevant features for a given application. While the initial paradigm was to have these applications run by a server hosted processor, recent advances in microelectronics provide hardware accelerators with an efficient ratio between computation and energy consumption, enabling the implementation of artificial intelligence algorithms ‘at the edge’. In this way only the meaningful and useful data are transmitted to the end-user, minimising the required data bandwidth, and reducing the latency with respect to the cloud computing model. In recent years, European Space Agency is promoting the development of disruptive innovative technologies on-board Earth Observation missions. In this field, the most advanced experiment to date is the Φ-sat-1, which has demonstrated the potential of Artificial Intelligence as a reliable and accurate tool for cloud detection on-board a hyperspectral imaging mission. The activities involved included demonstrating the robustness of the Intel Movidius Myriad 2 hardware accelerator against ionising radiation, developing a Cloudscout segmentation neural network, run on Myriad 2, to identify, classify, and eventually discard on-board the cloudy images, and assessing of the innovative Hyperscout-2 hyperspectral sensor. This mission represents the first official attempt to successfully run an AI Deep Convolutional Neural Network (CNN) directly inferencing on a dedicated accelerator on-board a satellite, opening the way for a new era of discovery and commercial applications driven by the deployment of on-board AI.

Several total and partial photoionization cross section calculations, based on both theoretical and empirical approaches, are quantitatively evaluated with statistical analyses using a large collection of experimental data retrieved from the literature to identify the state of the art for modeling the photoelectric effect in Monte Carlo particle transport. Some of the examined cross section models are available in general purpose Monte Carlo systems, while others have been implemented and subjected to validation tests for the first time to estimate whether they could improve the accuracy of particle transport codes. The validation process identifies Scofield's 1973 non-relativistic calculations, tabulated in the Evaluated Photon Data Library(EPDL), as the one best reproducing experimental measurements of total cross sections. Specialized total cross section models, some of which derive from more recent calculations, do not provide significant improvements. Scofield's non-relativistic calculations are not surpassed regarding the compatibility with experiment of K and L shell photoionization cross sections either, although in a few test cases Ebel's parameterization produces more accurate results close to absorption edges. Modifications to Biggs and Lighthill's parameterization implemented in Geant4 significantly reduce the accuracy of total cross sections at low energies with respect to its original formulation. The scarcity of suitable experimental data hinders a similar extensive analysis for the simulation of the photoelectron angular distribution, which is limited to a qualitative appraisal., To be published in IEEE Trans. Nucl. Sci

We introduce a theoretical framework for the calculation of uncertainties affecting observables produced by Monte Carlo particle transport, which derive from uncertainties in physical parameters input into simulation. The theoretical developments are complemented by a heuristic application, which illustrates the method of calculation in a streamlined simulation environment., 12 pages, 9 figures

A system for in-vivo tracking of 192Ir source during high dose rate or pulsed dose rate brachytherapy treatment was built using 1 mm thick silicon pad detectors as image sensors and knife-edge lead pinholes as collimators. With source self-images obtained from a dual-pinhole system, location of the source could be reconstructed in three dimensions in real time. The system was tested with 192Ir clinical source (kerma rate in air at 1 m 2.38 Gy/h) in air and plexi-glass phantom. The locations of the source were tracked from a distance of 40 cm in a field of view of 20 × 20 × 20 cm3. Reconstruction precision, defined as the average distance between true and reconstructed source positions, with data collected in less than 1 s with 22 GBq 192Ir source was about 5 mm. The reconstruction precision was in our case mainly limited by imperfect alignment of detectors and pinholes. With perfect alignment the statistical error would allow precision of about 1 mm which could further be improved with larger detector placed at larger distance from the pinhole. However already the modest precision of few millimeters is sufficient for in-vivo detection of larger deviations from planned treatment caused by various misadministrations or malfunctioning of the brachytherapy treatment apparatus. Usage of silicon detectors offers a possibility for building a compact device which could be used as an independent online quality assurance system. In this paper details about sensors, readout system and reconstruction algorithm are described. Results from measurements with clinical source are presented.