C-2 Innovations (United States)

The Neurodevelopmental Ecological Screening Tool (NEST) is a new instrument to screen children for developmental challenges. This article describes the validation of the NEST neurodevelopmental domain. Data were collected from a nationwide purposely restricted sample of caregivers of children aged 3–5 years ( n = 231) living in poverty and experiencing homelessness. We used Rasch-based Rating Scale Models to select items with good fit. Cronbach’s alpha was used to measure the internal consistency validity of the entire neurodevelopmental domain. Construct validity and dimensional structure were obtained using confirmatory factor analysis. Interclass correlations were used to measure the test–retest reliability of the subdomains. Neurodevelopmental Ecological Screening Tool scores were compared to results on clinician-administered gold standard measures for a subsample ( n = 48). The neurodevelopmental domain score represents a single overarching risk construct with some variance attributable to distinct developmental constructs and validly and reliably identifies a child’s level of developmental risk.

Resolution in deep convolutional neural networks (CNNs) is typically bounded by the receptive field size through filter sizes, and subsampling layers or strided convolutions on feature maps. The optimal resolution may vary significantly depending on the dataset. Modern CNNs hard-code their resolution hyper-parameters in the network architecture which makes tuning such hyper-parameters cumbersome. We propose to do away with hard-coded resolution hyper-parameters and aim to learn the appropriate resolution from data. We use scale-space theory to obtain a self-similar parametrization of filters and make use of the N-Jet: a truncated Taylor series to approximate a filter by a learned combination of Gaussian derivative filters. The parameter σ of the Gaussian basis controls both the amount of detail the filter encodes and the spatial extent of the filter. Since σ is a continuous parameter, we can optimize it with respect to the loss. The proposed N-Jet layer achieves comparable performance when used in state-of-the art architectures, while learning the correct resolution in each layer automatically. We evaluate our N-Jet layer on both classification and segmentation, and we show that learning σ is especially beneficial when dealing with inputs at multiple sizes.

The term non-zonal approach is applied in Go4Hybrid to refer to hybrid RANS-LES methods in which the model, not the user, defines the regions in which RANS and LES modes are active.

Multi-modal biometric verification systems are in active development and show impressive performance nowadays. However, such systems need additional protection from spoofing attacks. In our paper we present full pipeline of anti-spoofing method (based on our previous work) for bimodal audiovisual verification system. This method allows to evaluate parameters of quality for a sequence of face images during a verification process. Based on this parameters it’s decided whether the data is suitable for processing by the standard method (fiducial points based audiovisual liveness detection, FALD). If the quality of data is not sufficient, then our system switches to a new algorithm (svm-based audiovisual liveness detection, SALD), which provides less protection quality, but is able to operate when FALD is unsuitable. To improve the quality of the FALD algorithm we have collected the special dataset. This dataset allows to get better reliability of the algorithm for searching of fiducial points on the user’s face image. Tests show that developed system can significantly improve the quality of anti-spoofing protection versus our previous work.

In this paper we present the latest improvements to the Russian spontaneous speech recognition system developed in Speech Technology Center (STC). Significant word error rate (WER) reduction was obtained by applying hypothesis rescoring with sophisticated language models. These were the Recurrent Neural Network Language Model and regularized Long-Short Term Memory Language Model. For acoustic modeling we used the deep neural network (DNN) trained with speaker-dependent bottleneck features, similar to our previous system. This DNN was combined with the deep Bidirectional Long Short-Term Memory acoustic model by the use of score fusion. The resulting system achieves WER of 16.4 %, with an absolute reduction of 8.7 % and relative reduction of 34.7 % compared to our previous system result on this test set.

Purpose: to investigate the clinical feasibility of off-line PET/CT for in-vivo treatment verification of proton therapy. Method and Materials: Two PMMA blocks and one inhomogeneous phantom consisting of PMMA and muscle and bone equivalent slabs were irradiated with one or two orthogonal SOBP proton fields (8×8 cm2 aperture, 10 cm modulation and 15 or 16 cm range in water). The targets were imaged using a PET/CT (CPS/Siemens Biograph Sensation 16) scanner within 20 min after irradiation. At first, a high dose of 8 Gy was delivered and 1 h listmode acquisition was performed to investigate image quality based on counting statistics in variable time framesets. In the other studies a maximum dose of 2 Gy was applied and acquisition was limited to 20 min to mimic realistic therapeutic cases. The amount and spatial distribution of measured activity was compared to calculations based on the FLUKA Monte Carlo code and experimental cross-sections. Proton range was extracted from the analysis up to the second derivative of the activity distal edge. Isotopes were identified from decay time analysis. Results: The shape of the irradiated field and the range-correlated activity distal edge could be imaged with sufficient accuracy after therapeutic doses, despite the delay between irradiation and scan. In PMMA, maximum 11C activation (0.9±0.1 kBq/Gy/ml after 1.0–1.6 Gy/min irradiation) and distal edge position agree within 2% and 1% with calculations, respectively. Besides 11C, minor amounts of 15O and 13N were identified at the beginning of acquisition. Conclusion: This feasibility study indicates the potential of off-line PET/CT for range and field position verification in proton therapy. In addition to PET alone, PET/CT provides information on possible anatomical changes during fractionated radiotherapy. Clinical patient studies addressing the accuracy and possible limitations due to perfusion are planned. If available, preliminary clinical results will be presented.

This paper explores fast reconstruction strategies for 3D time-of-flight (TOF) positron emission tomography (PET), based on 2D data rebinning. Starting from pre-corrected 3D TOF data, a rebinning algorithm estimates for each transaxial slice the 2D TOF sinogram that would have been acquired by a single-ring scanner. The rebinned sinograms can then be reconstructed using any algorithm for 2D TOF reconstruction. We introduce TOF-FORE, an approximate rebinning algorithm obtained by extending the Fourier rebinning method for non-TOF data. In addition, we identify two partial differential equations that must be satisfied by consistent 3D TOF data, and use them to derive exact rebinning algorithms and to characterize the degree of the approximation in TOF-FORE. Numerical simulations demonstrate that TOF-FORE is more accurate than two different TOF extensions of the single-slice rebinning method, and suggest that TOF-FORE will be a valuable tool for practical TOF PET in the range of axial apertures and time resolutions typical of current scanners.

The high resolution research tomograph (HRRT) is a brain-dedicated scanner manufactured by CPS Innovations using LSO panel detectors. Transmission is measured using a /sup 137/Cs point source, which is moved axially and rotated to cover the FOV. The point source is collimated axially and transaxially to illuminate only a few planes on the heads opposite to the point source location. Pseudo-coincidence events are generated using a given crystal and the source location. The transmission system was previously validated for cold transmissions. For post-injection (hot) transmission, it is not possible to eliminate the emission contamination by raising the lower energy threshold. Since real mock scan is unpractical on HRRT and fake mock scan requires additional data, we developed a new technique to simultaneously measure the transmission and the mock scans. The technique uses a virtual source, axially located at a distance equal to the half of the axial FOV and illuminates a fan separated from the real transmission fan. We validated the shifted-mock scan technique by comparing it to real mock scan one with a /sup 68/Ge phantom and examined its effectiveness with a hot 20 cm phantom filled with /sup 18/F decaying over several half-lives. Local residual bias in /spl mu/-map was attributed to transmission scatter and corrected by using partial segmentation in the MAP-TR algorithm, /spl mu/-maps from cold and hot transmissions were compared on several clinical patients and a Hoffman brain phantom for which their influence on emission quantification was studied.