Search

Patients with obstructive sleep apnea (OSA) show brain injury in sites responsible for autonomic, cognitive, and respiratory functions. Brain changes in OSA may vary with disease severity as assessed by the apnea-hypopnea index (AHI), which does not provide information about the apnea depth and length in contrast to oxygen desaturation. Although significant associations with brain injury and AHI are known in OSA, it is unclear whether AHI or the extent of oxygen desaturations better correlate with brain damage. We evaluated associations between brain changes, AHI, and oxygen desaturation using diffusion tensor imaging-based measures.We acquired diffusion tensor imaging data from 19 patients with OSA using a 3.0-Tesla MRI scanner and calculated, normalized, and smoothed mean, axial, and radial diffusivity maps that were used for correlations between brain changes, oxygen desaturation, and AHI values.Positive correlations with extent of injury (mean, axial, and radial diffusivity values) and AHI appeared in the frontal areas, cingulate and insula, amygdala, hippocampus, and basal pons, and negative associations emerged in the putamen, internal-capsule, globus-pallidus, and cerebellar cortices. Regional diffusivity values and oxygen desaturation showed positive correlations in the cingulate, frontal, putamen, and cerebellar sites, and negative relationships in several areas, including the occipital cortex.Patients with OSA show negative and positive correlations, indicated by increased and decreased diffusivity values, resulting from chronic and acute changes in those areas. The extent of injury in OSA partially depends on the extent of AHI and oxygen desaturation, with the effects representing continued development from acute to chronic processes.Sahib A, Roy B, Kang D, Aysola RS, Wen E, Kumar R. Relationships between brain tissue damage, oxygen desaturation, and disease severity in obstructive sleep apnea evaluated by diffusion tensor imaging.

The availability of vast video collections and the accuracy of ML models has generated significant interest in video analytics systems. Since naively processing all frames using expensive models is impractical, researchers have proposed optimizations such as selectively using faster but less accurate models to replace or filter frames for expensive models. However, these optimizations are difficult to apply on queries with multiple predicates and models, as users must manually explore a large optimization space. Without significant systems expertise or time investment, an analyst may manually create an execution plan that is unnecessarily expensive and/or terribly inaccurate. We propose Relational Hints , a declarative interface that allows users to suggest ML model relationships based on domain knowledge. Users can express two key relationships: when a model can replace another (CAN REPLACE) and when a model can be used to filter frames for another (CAN FILTER). We aim to design an interface to express model relationships informed by domain specific knowledge and define the constraints by which these relationships hold. We then present the VIVA video analytics system that uses relational hints to optimize SQL queries on video datasets. VIVA automatically selects and validates the hints applicable to the query, generates possible query plans using a formal set of transformations, and finds the best performance plan that meets a user's accuracy requirements. VIVA relieves users from rewriting and manually optimizing video queries as new models become available and execution environments evolve. We evaluate VIVA implemented on top of Spark and show that hints improve performance up to 16.6X without sacrificing accuracy.

While deep neural networks (DNNs) are an increasingly popular way to query large corpora of data, their significant runtime remains an active area of research. As a result, researchers have proposed systems and optimizations to reduce these costs by allowing users to trade off accuracy and speed. In this work, we examine end-to-end DNN execution in visual analytics systems on modern accelerators. Through a novel measurement study, we show that the preprocessing of data (e.g., decoding, resizing) can be the bottleneck in many visual analytics systems on modern hardware. To address the bottleneck of preprocessing, we introduce two optimizations for end-to-end visual analytics systems. First, we introduce novel methods of achieving accuracy and throughput trade-offs by using natively present, low-resolution visual data. Second, we develop a runtime engine for efficient visual DNN inference. This runtime engine a) efficiently pipelines preprocessing and DNN execution for inference, b) places preprocessing operations on the CPU or GPU in a hardware- and input-aware manner, and c) efficiently manages memory and threading for high throughput execution. We implement these optimizations in a novel system, Smol, and evaluate Smol on eight visual datasets. We show that its optimizations can achieve up to 5.9X end-to-end throughput improvements at a fixed accuracy over recent work in visual analytics.

Systems for ML inference are widely deployed today, but they typically optimize ML inference workloads using techniques designed for conventional data serving workloads and miss critical opportunities to leverage the statistical nature of ML. In this demo, we present Willump, an optimizer for ML inference that introduces statistically-motivated optimizations targeting ML applications whose performance bottleneck is feature computation. Willump automatically cascades feature computation for classification queries: Willump classifies most data inputs using only high-value, low-cost features selected by a cost model, improving query performance by up to 5 x without statistically significant accuracy loss. In this demo, we use interactive and easily-downloadable Jupyter notebooks to show VLDB attendees which applications Willump can speed up, how to use Willump, and how Willump produces such large performance gains.

PURPOSE To validate healthcare claim-based algorithms for neurodevelopmental disorders (NDD) in children using medical records as the reference. METHODS Using a clinical data warehouse of patients receiving outpatient or inpatient care at two hospitals in Boston, we identified children (≤14 years between 2010 and 2014) with at least one of the following NDDs according to claims-based algorithms: autism spectrum disorder/pervasive developmental disorder (ASD), attention deficit disorder/other hyperkinetic syndromes of childhood (ADHD), learning disability, speech/language disorder, developmental coordination disorder (DCD), intellectual disability, and behavioral disorder. Fifty cases per outcome were randomly sampled and their medical records were independently reviewed by two physicians to adjudicate the outcome presence. Positive predictive values (PPVs) and 95% confidence intervals (CIs) were calculated. RESULTS PPVs were 94% (95% CI, 83%-99%) for ASD, 88% (76%-95%) for ADHD, 98% (89%-100%) for learning disability, 98% (89%-100%) for speech/language disorder, 82% (69%-91%) for intellectual disability, and 92% (81%-98%) for behavioral disorder. A total of 19 of the 50 algorithm-based cases of DCD were confirmed as severe coordination disorders with functional impairment, with a PPV of 38% (25%-53%). Among the 31 false-positive cases of DCD were 7 children with coordination deficits that did not persist throughout childhood, 7 with visual-motor integration deficits, 12 with coordination issues due to an underlying medical condition and 5 with ADHD and at least one other severe NDD. CONCLUSIONS PPVs were generally high (range: 82%-98%), suggesting that claims-based algorithms can be used to study NDDs. For DCD, additional criteria are needed to improve the classification of true cases.

Pharmaceuticals 14(8), 818 (2021). doi:10.3390/ph14080818 special issue: "Special Issue "Targets, Tracers and Translation, Part 2 - New Horizons in Radiopharmaceutical Development" / Special Issue Editor: Prof. Dr. Gerald Reischl, Guest Editor", Published by MDPI, Basel

Researchers and industry analysts are increasingly interested in computing aggregation queries over large, unstructured datasets with selective predicates that are computed using expensive deep neural networks (DNNs). As these DNNs are expensive and because many applications can tolerate approximate answers, analysts are interested in accelerating these queries via approximations. Unfortunately, standard approximate query processing techniques to accelerate such queries are not applicable because they assume the result of the predicates are available ahead of time. Furthermore, recent work using cheap approximations (i.e., proxies) do not support aggregation queries with predicates. To accelerate aggregation queries with expensive predicates, we develop and analyze a query processing algorithm that leverages proxies (ABAE). ABAE must account for the key challenge that it may sample records that do not satisfy the predicate. To address this challenge, we first use the proxy to group records into strata so that records satisfying the predicate are ideally grouped into few strata. Given these strata, ABAE uses pilot sampling and plugin estimates to sample according to the optimal allocation. We show that ABAE converges at an optimal rate in a novel analysis of stratified sampling with draws that may not satisfy the predicate. We further show that ABAE outperforms on baselines on six real-world datasets, reducing labeling costs by up to 2.3X.

Over the past 20 years, 68Ga-labelled radiopharmaceuticals have become an important part in clinical routine. However, the worldwide supply with 68Ge/68Ga generators is limited as well as the number of patient doses per batch of 68Ga radiopharmaceutical. In the recent years, a new technique appeared, making use of the ease of aqueous labelling via chelators as with 68Ga but using 18F instead. This technique takes advantage of the strong coordinative bond between aluminium and fluoride, realized in the aqueous cation [Al18F]2+. Most applications to date make use of one-pot syntheses with free Al(III) ions in the system. In contrast, we investigated the labelling approach split into two steps: generating the Al-bearing precursor in pure form and using this Al compound as a precursor in the labelling step with aqueous [18F]fluoride. Hence, no free Al3+ ions are present in the labelling step. We investigated the impact of parameters: temperature, pH, addition of organic solvent, and reaction time using the model chelator NH2-MPAA-NODA. With optimized parameters we could stably achieve a 80% radiochemical yield exerting a 30-min reaction time at 100 °C. This technique has the potential to become an important approach in radiopharmaceutical syntheses.

A liquid film with two distinct contact lines underneath bubbles during boiling on structured surfaces has been discovered., Enhancing critical heat flux (CHF) during boiling with structured surfaces has received much attention because of its important implications for two-phase flow. The role of surface structures on bubble evolution and CHF enhancement remains unclear because of the lack of direct visualization of the liquid- and solid-vapor interfaces. Here, we use high-magnification in-liquid endoscopy to directly probe bubble behavior during boiling. We report the previously unidentified coexistence of two distinct three-phase contact lines underneath growing bubbles on structured surfaces, resulting in retention of a thin liquid film within the structures between the two contact lines due to their disparate advancing velocities. This finding sheds light on a previously unidentified mechanism governing bubble evolution on structured surfaces, which has notable implications for a variety of real systems using bubble formation, such as thermal management, microfluidics, and electrochemical reactors.

Sub-micron pixels have been widely adopted in recent CMOS image sensors to implement high resolution cameras in small form factors, i.e. slim mobile-phones. Even with shrinking pixels, customers demand higher image quality, and the pixel performance must remain comparable to that of the previous generations. Conventionally, to suppress the optical crosstalk between pixels, a metal grid has been used as an isolation structure between adjacent color filters. However, as the pixel size continues to shrink to the sub-micron regime, an optical loss increases because the focal spot size of the pixel’s microlens does not downscale accordingly with the decreasing pixel size due to the diffraction limit: the light absorption inevitably occurs in the metal grid. For the first time, we have demonstrated a new lossless, dielectric-only grid scheme. The result shows 29 % increase in sensitivity and +1.2-dB enhancement in Y-SNR when compared to the previous hybrid metal-and-dielectric grid.

Study objective To provide a contemporary medicolegal analysis of claims brought against anesthesia providers in the United States related to neuraxial blocks for surgery and obstetrics. Design In this retrospective analysis, we analyzed closed claims data from the Controlled Risk Insurance Company (CRICO) Comparative Benchmarking System (CBS) database between 2007 and 2016. Setting Closed claims from inpatient and outpatient settings related to neuraxial anesthesia for surgical procedures and obstetrics. Patients Forty-five claims were identified for analysis. These patients underwent a variety of surgical procedures, included both children and adults, and with ages ranging from 6 to 82. Interventions Patients receiving neuraxial anesthesia (spinals, epidurals) for surgery or obstetrics. Measurements Data collected includes patient demographics, alleged injury type/severity, surgical specialty, likely contributors to the alleged damaging event, and case outcome. Some of the data were drawn directly from coded variables in the CRICO database, and some were gathered from narrative case summaries. Main results Settlement payments were made in 20% of claims. Reported adverse outcomes ranged from temporary minor to permanent major injuries. Most closed claims were classified as permanent minor injuries. The greatest number of claims involved residual weakness and radiculopathy resulting from epidurals. The largest contributing factor to these injuries was noted to be “Technical Knowledge/Performance” of the anesthesia provider followed by “Missing or Documentation Error.” Over half of the claims arose from obstetric patients (31%) and patients undergoing orthopedic surgery (27%). Conclusions Patients with pre-existing radiculopathy or comorbidities may warrant more thorough informed consent about the increased risk of injury. Additionally, prompt follow-up, monitoring, and documentation of post-operative symptoms, such as weakness or radiculopathy, are crucial for improving patient safety and satisfaction. More timely communication with the patient and the surgical team regarding residual neurologic symptoms is important for earlier diagnosis of injury.

Researchers have proposed hardware, software, and algorithmic optimizations to improve the computational performance of deep learning. While some of these optimizations perform the same operations faster (e.g., increasing GPU clock speed), many others modify the semantics of the training procedure (e.g., reduced precision), and can impact the final model's accuracy on unseen data. Due to a lack of standard evaluation criteria that considers these trade-offs, it is difficult to directly compare these optimizations. To address this problem, we recently introduced DAWNBENCH, a benchmark competition focused on end-to-end training time to achieve near-state-of-the-art accuracy on an unseen dataset-a combined metric called time-to-accuracy (TTA). In this work, we analyze the entries from DAWNBENCH, which received optimized submissions from multiple industrial groups, to investigate the behavior of TTA as a metric as well as trends in the best-performing entries. We show that TTA has a low coefficient of variation and that models optimized for TTA generalize nearly as well as those trained using standard methods. Additionally, even though DAWNBENCH entries were able to train ImageNet models in under 3 minutes, we find they still underutilize hardware capabilities such as Tensor Cores. Furthermore, we find that distributed entries can spend more than half of their time on communication. We show similar findings with entries to the MLPERF v0.5 benchmark.

Background: Closed malpractice claims can provide insight into low-frequency adverse events in many areas of perioperative and chronic pain care. Over the last decade, there have been changes in surgical and regional anesthetic practice, likely impacting adverse event patterns. Given the wide variability and low frequency of complications associated with peripheral nerve blocks, the study of closed malpractice claims offers an opportunity to examine adverse events, and the patient, technical, and provider factors that led to the claim. Knowledge gained from examination of closed claims has already resulted in multiple improvements in processes of care and patient safety. Objectives: An investigation of the factors that contributed to medicolegal claims against anesthesia providers related to peripheral nerve blocks. Study Design: Retrospective analysis. Setting: Inpatient and outpatient surgery facilities. Methods: The Comparative Benchmarking System database is a medical liability database that contains more than 400,000 malpractice claims from more than 400 academic and communitybased institutions accounting for over 30% of malpractice claims in the United States. The present investigation reviewed all (n = 113) available closed malpractice claims related to regional anesthesia (RA) in surgical patients closed between 2006 and 2016, and investigated factors that may have contributed to patient injury, including type of nerve block, type of surgery, nerves injured, resulting neurologic deficits, and potential factors contributing to the injury. Results: Our data analyzed 62 claims related to RA and showed that most closed claims were classified as permanent minor injuries. The greatest number of claims were for brachial plexus injuries associated with interscalene blocks performed for shoulder or rotator cuff repairs. Femoral and sciatic nerve blocks with resulting lower extremity injuries were the most common nerve blocks resulting in payment. The largest contributing factor to these injuries was noted to be “Technical Knowledge/Performance” of the regionalist followed by “Pre-existing Injury/Radiculopathy.” Symptom onset from these claims was most likely to be delayed with the leading initial presenting symptom being paresthesia. Limitations: It is difficult to establish cause-effect relationship, and the small sample size limits the ability to detect clinical differences and associations with specific comorbidities or techniques. There was also limited information related to regional anesthetic techniques and medications used that would have helped explore further relationships between the procedure and cause for litigation. Conclusions: There remains significant room for risk reduction in regional anesthetic practice. Patterns based on the analysis of closed claims show that interscalene blocks are the most common peripheral nerve block resulting in litigation, even when compared with other blocks involving the brachial plexus. Furthermore, patients with existing nerve injury/radiculopathy may also warrant alternative techniques or greater emphasis during informed consent on the increased risk of injury. As most of the presenting symptoms associated with claims are delayed, an opportunity for improvement in postregional care may be better communication with patients following discharge to discuss their postoperative recovery. Key words: Regional, pain, anesthesia, complications, closed claims, liability, nerve, block, injury

Objectives To determine if the Patient-Reported Outcome Measurement Information System (PROMIS) physical function, pain interference, self-efficacy, and global rating of normal function (GRNF) scales are able to accurately characterize a patient’s acceptable symptom state (PASS). Design A cross-sectional analysis, using receiver operator curves and chi-square analysis to explore criteria to determine thresholds (80% and 95% sensitivity/specificity) for PASS that are applicable to PROMIS and GRNF scales. Setting Phone survey after primary care. Participants Patients (N=94) attending primary care for musculoskeletal problems. Interventions Not applicable. Main Outcomes Measures Accuracy and proportion of patients classified as PASS Yes or No. Results Receiver operator curve analysis showed significant area under the curve (AUC) values for each PROMIS scale (AUC>.72) and the GRNF rating (AUC=.74). Identified PROMIS thresholds suggested PASS was achieved when scores were at or slightly worse than the US population average. A score of ≥7 and ≤4 characterized patients that were PASS Yes and No, respectively, on the GRNF rating. A moderate (80%) specificity/sensitivity criteria yielded 72.3%-73.5% accuracy for a majority of participants (>69.9%). Conclusion This analysis suggests the PROMIS and GRNF scales are able to characterize PASS status with moderate accuracy (∼70%) for a large portion of patients (∼70%). New to this study is the association of self-efficacy with PASS status. PROMIS scales at or slightly worse than the US population average characterized PASS status.