FARE2017 WINNERS Sorted by Institute/Center
Total Page:16
File Type:pdf, Size:1020Kb
FARE2017 WINNERS Sorted By Institute/Center NIH Clinical Center (CC) Zachary Lerner Postdoctoral FellowPostdoctoral Fellow Clinical and Translational Research A Robotic Exoskeleton to Treat Crouch Gait from Cerebral Palsy: Design and Initial Clinical Evaluation Cerebral palsy (CP) is the most prevalent childhood physical disability and adversely affects walking and other motor abilities. Crouch gait, a pathological walking pattern characterized by excessive knee flexion, is one of the most common gait disorders observed in children with CP. Effective treatment of crouch during childhood is critical to maintain mobility into adulthood, yet current interventions do not adequately eliminate or alleviate crouch in most patients. Wearable robotic exoskeletons have the potential to improve crouch gait by providing on demand assistance during walking. However, no exoskeletons suitable to treat children are commercially available, and no evidence exists regarding the feasibility or efficacy of utilizing motorized assistance to alleviate knee flexion from crouch gait. Enhanced knowledge of neuromuscular adaptations to powered assistance in children with crouch gait is needed to optimize this treatment approach. To meet these needs, we developed the first lower- extremity exoskeleton intended to treat crouch gait by providing powered knee extension assistance at different phases of the gait cycle. We evaluated the effects of powered knee extension assistance on knee motion, and knee flexor and extensor muscle activity in four children with crouch gait from CP. The exoskeleton was effective in reducing crouch in three of the four participants compared to their baseline gait pattern. The reduction in crouch was clinically significant (greater than 10 degrees) in two of the participants. Knee extensor activity was maintained during early stance in all of the participants during walking with the exoskeleton which is a desirable outcome because we want the device to supplement voluntary control, not replace it. In two of the participants, knee extensor activity in mid-stance represented a more normal, well-modulated pattern. Modest increases in knee flexor activity were also exhibited. Therefore, additional training focusing on reducing knee flexor activity may lead to further improvements. In summary, we demonstrated the promise of a wearable robotic exoskeleton as a potential treatment for individuals with crouch gait. Our results provide novel insights into motor control strategies for individuals with CP and may enhance understanding of the neuromuscular causes underlying crouch gait. While currently a laboratory-based intervention, the ultimate goal is to prescribe the exoskeleton as a device for long-term rehabilitation. NIH Clinical Center (CC) Kristina Brooks Postdoctoral FellowPostdoctoral Fellow Pharmacology and Toxicology/Environmental Health Cobicistat Increases the Effects of Dabigatran on Thrombin Time in Healthy Volunteers Background: Dabigatran etexilate (DE) is an oral direct thrombin inhibitor and substrate of intestinal Permeability-glycoprotein (P-gp) and renal multidrug and toxin extrusion-1 (MATE-1) transporters. Cobicistat (COBI) is a pharmacokinetic (PK) enhancer and inhibitor of P-gp and renal MATE-1 transporters, which may result in increased DE exposure when coadministered. Using thrombin time (TT), we sought to characterize the pharmacodynamic (PD) effects of COBI on DE, and if this potential interaction may be mitigated by separated administration. Methods: This was a single-center, open- label, fixed sequence, intra-subject study conducted in healthy volunteers. The study was comprised of 3 phases: (1) DE 150 mg x1 alone (day 0), (2) DE 150 mg x1 two hours prior to COBI 150 mg (day 19±1), and (3) DE 150 mg x1 with COBI 150 mg (day 26±1). Subjects underwent a 5-day washout period following phase 1, followed by COBI 150 mg once daily throughout Phases 2 and 3 (days 5 - 26±1). Blood was collected at 0, 0.5, 1, 2, 3, 4, 6, 8, 12, and 24 hours post-dose on DE dosing days. TT was determined using STA®-Thrombin reagent (Diagnostica Stago, Asnières-sur-Seine, France). Noncompartmental methods were used to derive DE PD parameters (TT area-under-the-effect-curve [AUEC] and thrombin time at 24 hours [TT-last]). Geometric mean ratios (GMR) with 90% confidence intervals [CI] were compared between phases and p-values were calculated using a two-tailed paired t-test. Results: A total of 18 subjects were enrolled, with 16 completing the study and 2 without phase 3 data due to lack of study compliance. There was a 30% and 33% increase in AUEC GMR between Phase 2 and Phase 1 (p below 0.0001, 90% CI [1.21, 1.39]), and Phase 3 and Phase 1 (p below 0.0001, 90% CI 1.22-1.44), respectively. Significant increases in TT-last GMR of 46% and 51% were also observed between Phase 2 and Phase 1 (p below 0.0001, 90% CI [1.30, 1.61]) and Phase 3 and Phase 1 (p below 0.001, 90% CI [1.24, 1.78]), respectively. No significant differences were noted between Phase 3 and Phase 2 in GMR of AUEC (p=0.551, 90% CI [0.95, 1.09]) or TT-last (p=0.595, 90% CI [0.89, 1.17]). Conclusions: COBI coadministration resulted in significant increases in TT AUEC and TT-last vs. DE alone. These effects were preserved despite separating COBI and DE administration by 2 hours, supporting the putative mechanisms of P-gp and MATE-1 inhibition by COBI. Further analyses of changes in DE PK-PD relationships are warranted. NIH Clinical Center (CC) Mingchen Gao Postdoctoral FellowPostdoctoral Fellow Radiology/Imaging/PET and Neuroimaging Multi-label Deep Convolutional Neural Networks for Holistic Interstitial Lung Disease Detection Holistically detecting interstitial lung diseases (ILD) using single CT images is a challenging but also important medical imaging problem. The difficulties lie on several aspects, which include the tremendous amount of variation in disease appearance, location, and configuration and also the expense required to obtain delicate pixel-level ILD annotations of large datasets for training. Finally, there are usually multiple diseases coexisting on the same patient, even on single CT slices. Beyond the basic approaches from most of the previous work, focusing on predicting a single ILD label to manually defined region of interest, we propose a multi-label deep regression model for holistic images. An end- to-end convolutional neural network (CNN) network is trained for multi-label image regression. The deep CNN regression model, which is inspired by the cortex of the brain, learns the deep image features and the final predictions to multi labels simultaneously. While CNNs are powerful tools, their feature learning strategy is not invariant to the spatial locations. To accommodate the large spatial variations of the ILD locations, the learned CNN features at different network depths are spatially aggregated and encoded through Fisher Vector (FV) method to turn them into location-invariant representations. The unordered features are then trained using a mutli-variate linear regressor to regress the numbers of ILD pixels or binary labels. The proposed algorithms are validated on a publicly available dataset of 533 patients, called Lung Tissue Research Consortium (LTRC) dataset, using five-fold cross-validation. Four most typical ILDs are investigated here, Ground Glass, Reticular, Honeycomb and Emphysema. There are 18883 slices in total for training and testing. There are 3368, 1606, 1247 and 2639 positive slices for each disease, respectively. In total there are 11677 healthy CT images, 5675 images with one disease, 1410 images with two diseases, 119 images with three diseases, and 2 images with four diseases. We achieved high area-under-curve (AUC) scores of 0.982, 0.972, 0.893 and 0.993 for Ground Glass, Reticular, Honeycomb and Emphysema, respectively. As such, our work represents an important step forward in providing clinically effective ILD detection. NIH Clinical Center (CC) Xiaosong Wang Visiting FellowVisiting Fellow Radiology/Imaging/PET and Neuroimaging Unsupervised Category Discovery via Looped Deep Pseudo-Task Optimization Using a Large Scale Radiology Image Database Obtaining semantic labels on a large scale radiology image database (215,786 key images from 61,845 unique patients) is a prerequisite yet bottleneck to train highly effective deep Convolutional Neural Network (CNN) models for image recognition. Nevertheless, conventional means of collecting image labels (e.g. Google image search followed by crowd-sourcing are not applicable due to 1) the formidable difficulties of medical annotation tasks for clinically untrained annotators, 2) unavailability of a high quality or large capacity medical image search engine. On the other hand, even for well-trained radiologists, this type of “assigning labels to images” task is not aligned with their regular diagnostic routine work so that drastic inter-observer variations or inconsistency may be demonstrated. In this project, we proposed a Looped Deep Pseudo-task Optimization (LDPO) procedure for automatic category discovery (auto-annotation) of visually coherent and clinically semantic (concept) clusters. Our system can be initialized by domain-specific (CNN trained on radiology images and text report derived labels) or generic (ImageNet based) CNN models. Afterwards, a sequence of pseudo-tasks are exploited by the looped deep image feature clustering (to refine image labels) and deep CNN training/classification using new labels (to obtain more task representative deep features). Our method