Laustenemery1383

Attention based end-to-end speech synthesis achieves better performance in both prosody and quality compared to the conventional "front-end"-"back-end" structure. But training such end-to-end framework is usually time-consuming because of the use of recurrent neural networks. #link# To enable parallel calculation and long-range dependency modeling, a solely self-attention based framework named Transformer is proposed recently in the end-to-end family. However, it lacks position information in sequential modeling, so that the extra position representation is crucial to achieve good performance. Besides, the weighted sum form of self-attention is conducted over the whole input sequence when computing latent representation, which may disperse the attention to the whole input sequence other than focusing on the more important neighboring input states, resulting in generation errors. In this paper, we introduce two localness modeling methods to enhance the self-attention based representation for speech synthesis, which maintain the abilities of parallel computation and global-range dependency modeling in self-attention while improving the generation stability. We systematically analyze the solely self-attention based end-to-end speech synthesis framework, and unveil the importance of local context. Then we add the proposed relative-position-aware method to enhance local edges and experiment with different architectures to examine the effectiveness of localness modeling. In order to achieve query-specific window and discard the hyper-parameter of the relative-position-aware approach, we further conduct Gaussian-based bias to enhance localness. Experimental results indicate that the two proposed localness enhanced methods can both improve the performance of the self-attention model, especially when applied to the encoder part. And the query-specific window of Gaussian bias approach is more robust compared with the fixed relative edges. PURPOSE To assess out-of-field doses in radiotherapy treatments of paediatric patients, using Monte Carlo methods to implement a new model of the linear accelerator validated against measurements and developing a voxelized anthropomorphic paediatric phantom. METHODS CT images of a physical anthropomorphic paediatric phantom were acquired and a dosimetric planning using a TPS was obtained. The CT images were used to perform the voxelization of the physical phantom using the ImageJ software and later implemented in MCNP. In order to validate the Monte Carlo model, dose measurements of the 6 MV beam and Linac with 120 MLC were made in a clinical setting, using ionization chambers and a water phantom. Afterwards TLD measurements in the physical anthropomorphic phantom were performed in order to assess the out-of-field doses in the eyes, thyroid, c-spine, heart and lungs. RESULTS The Monte Carlo model was validated for in-field and out-of-field doses with average relative differences below 3%. The average relative differences between TLD measurements and Monte Carlo is 14,3% whilst the average relative differences between TLD and TPS is 55,8%. Moreover, organs up to 22.5 cm from PTV center show TLD and MCNP6 relative differences and TLD and TPS relative differences up to 21.2% and 92.0%, respectively. CONCLUSIONS Our study provides a novel model that could be used in clinical research, namely in dose evaluation outside the treatment fields. This is particularly relevant, especially in pediatric patients, for studying new radiotherapy treatment techniques, since it can be used to estimate the development of secondary tumours. PURPOSE The purpose of this study is to employ magnetic fluid hyperthermia simulations in the precise computation of Specific Absorption Rate functions -SAR(T)-, and in the evaluation of the predictive capacity of different SAR calculation methods. METHODS Magnetic fluid hyperthermia experiments were carried out using magnetite-based nanofluids. The respective SAR values were estimated through four different calculation methods including the initial slope method, the Box-Lucas method, the corrected slope method and the incremental analysis method (INCAM). A novel numerical model combining the heat transfer equations and the Navier-Stokes equations was developed to reproduce the experimental heating process. To address variations in heating efficiency with temperature, the expression of the power dissipation as a Gaussian function of temperature was introduced and the Levenberg-Marquardt optimization algorithm was employed to compute the function parameters and determine the function's effective branch within each measurement's temperature range. The power dissipation function was then reduced to the respective SAR function. RESULTS The INCAM exhibited the lowest relative errors ranging between 0.62 and 15.03% with respect to the simulations. SAR(T) functions exhibited significant variations, up to 45%, within the MFH-relevant temperature range. CONCLUSIONS The examined calculation methods are not suitable to accurately quantify the heating efficiency of a magnetic fluid. Numerical models can be exploited to effectively compute SAR(T) and contribute to the development of robust hyperthermia treatment planning applications. PURPOSE To assess the performance of a new optimization system, VOLO, for CyberKnife MLC-based SBRT plans in comparison with the existing Sequential optimizer. METHODS MLC-plans were created for 25 SBRT cases (liver, prostate, pancreas and spine) using both VOLO and Sequential. Monitor units (MU), delivery time (DT), PTV coverage, conformity (nCI), dose gradient (R50%) and OAR doses were used for comparison and combined to obtain a mathematical score (MS) of plan quality for each solution. MS strength was validated by changing parameter weights and by a blinded clinical plan evaluation. The optimization times (OT) and the average segment areas (SA) were also compared. RESULTS VOLO solutions offered significantly lower mean DT (-19%) and MU (-13%). OTX008 datasheet were below 15 min for VOLO, whereas for Sequential, values spanned from 8 to 160 min. SAs were significantly larger for VOLO on average 10 cm2 versus 7 cm2. VOLO optimized plans achieved a higher MS than Sequential for all tested parameter combinations. PTV coverage and OAR sparing were comparable for both groups of solutions.