Dense reinforcement learning for safety validation of autonomous vehicles.

One critical bottleneck that impedes the development and deployment of autonomous vehicles is the prohibitively high economic and time costs required to validate their safety in a naturalistic driving environment, owing to the rarity of safety-critical events1. Here we report the development of an intelligent testing environment, where artificial-intelligence-based background agents are trained to validate the safety performances of autonomous vehicles in an accelerated mode, without loss of unbiasedness. From naturalistic driving data, the background agents learn what adversarial manoeuvre to execute through a dense deep-reinforcement-learning (D2RL) approach, in which Markov decision processes are edited by removing non-safety-critical states and reconnecting critical ones so that the information in the training data is densified. D2RL enables neural networks to learn from densified information with safety-critical events and achieves tasks that are intractable for traditional deep-reinforcement-learning approaches. We demonstrate the effectiveness of our approach by testing a highly automated vehicle in both highway and urban test tracks with an augmented-reality environment, combining simulated background vehicles with physical road infrastructure and a real autonomous test vehicle. Our results show that the D2RL-trained agents can accelerate the evaluation process by multiple orders of magnitude (103 to 105 times faster). In addition, D2RL will enable accelerated testing and training with other safety-critical autonomous systems.

User perceptions of automated Truck-Mounted attenuators: Implications on work zone safety.

OBJECTIVE: Automated Truck-Mounted Attenuators (ATMAs) have the potential to improve work zone safety by removing the human driver out of a vehicle that is positioned in work zones to absorb impact from errant vehicles. However, this automated technology is expensive and can be detrimental to safety and project success if operated incorrectly (e.g., operating limitations and procedures not followed). Therefore, it is important to understand users' perceptions of ATMAs and how training can improve appropriate adoption of this technology. The objective of this study was to evaluate how work zone workers perceive the usefulness of and the capabilities of automation in Truck-Mounted Attenuators. METHODS: A survey study was conducted with 13 Department of Transportation (DOT) workers in Colorado and California. Each of the DOT workers in this study had some previous experience with the ATMA, either in real-world applications and/or formal training. The survey collected information on participant job specifications, experience with the ATMA, training received, trust in the ATMA, usability of the HMIs, and operating capabilities of the automation. RESULTS: Workers reported an overall positive acceptance of this technology. This was supported by their expectation that it would reduce crash severity; that there was a reasonable workload associated with operating procedures for the automation; and by their overall trust in the automation's reliability. However, workers noted concerns regarding their trust in the automation under various contexts, such as poor visibility and denser traffic volumes. Further, trust in the technology was greatest among workers with higher levels of ATMA training and longer experience working with the ATMA. CONCLUSIONS: This research presents a novel perspective on user acceptance of ATMA technology. These findings can help jurisdictions achieve the safety improvements that investment and deployment of automation in work zones offers, by identifying the disconnect between operators and technology.

Health Technology Assessment of a new water quality monitoring technology: Impact of automation, digitalization and remoteness in dialysis units.

BACKGROUND: Water quality monitoring at the dialysis units (DU) is essential to ensure an appropriate dialysis fluid quality and guarantee an optimal and safe dialysis treatment to patients. This paper aims to evaluate the effectiveness, economic and organizational impact of automation, digitalization and remote water quality monitoring, through a New Water Technology (NWT) at a hospital DU to produce dialysis water, compared to a Conventional Water Technology (CWT). METHODS: A before-and-after study was carried out at the Hospital Clínic Barcelona. Data on CWT was collected during 1-year (control) and 7-month for the NWT (case). Data on water quality, resource use and unit cost were retrospective and prospectively collected. A comparative effectiveness analysis on the compliance rate of quality water parameters with the international guidelines between the NWT and the CWT was conducted. This was followed by a cost-minimization analysis and an organizational impact from the hospital perspective. An extensive deterministic sensitivity analysis was also performed. RESULTS: The NWT compared to the CWT showed no differences on effectiveness measured as the compliance rate on international requirements on water quality (100% vs. 100%), but the NWT yielded savings of 3,599 EUR/year compared to the CWT. The NWT offered more data accuracy (daily measures: 6 vs. 1 and missing data: 0 vs. 20 days/year), optimization of the DU employees' workload (attendance to DU: 4 vs. 19 days/month) and workflow, through the remote and continuous monitoring, reliability of data and process regarding audits for quality control. CONCLUSIONS: While the compliance of international recommendations on continuous monitoring was performed with the CWT, the NWT was efficient compared to the CWT, mainly due to the travel time needed by the technical operator to attend the DU. These results were scalable to other economic contexts. Nonetheless, they should be taken with caution either when the NWT equipment/maintenance cost are largely increased, or the workforce involvement is diminished.

Automation, workers' skills and job satisfaction.

When industrial robots are adopted by firms in a local labor market, some workers are displaced and become unemployed. Other workers that are not directly affected by automation may however fear that these new technologies might replace their working tasks in the future. This fear of a possible future replacement is important because it negatively affects workers' job satisfaction at present. This paper studies the extent to which automation affects workers' job satisfaction, and whether this effect differs for high- versus low-skilled workers. The empirical analysis uses microdata for several thousand workers in Norway from the Working Life Barometer survey for the period 2016-2019, combined with information on the introduction of industrial robots in Norway from the International Federation of Robotics. Our identification strategy exploits variation in the pace of introduction of industrial robots in Norwegian regions and industries since 2007 to instrument workers' fear of replacement. The results indicate that automation in industrial firms in recent years have induced 40% of the workers that are currently in employment to fear that their work might be replaced by a smart machine in the future. Such fear of future replacement does negatively affect workers' job satisfaction at present. This negative effect is driven by low-skilled workers, which are those carrying out routine-based tasks, and who are therefore more exposed to the risks of automation.

Computational planning of the synthesis of complex natural products.

Training algorithms to computationally plan multistep organic syntheses has been a challenge for more than 50 years1-7. However, the field has progressed greatly since the development of early programs such as LHASA1,7, for which reaction choices at each step were made by human operators. Multiple software platforms6,8-14 are now capable of completely autonomous planning. But these programs 'think' only one step at a time and have so far been limited to relatively simple targets, the syntheses of which could arguably be designed by human chemists within minutes, without the help of a computer. Furthermore, no algorithm has yet been able to design plausible routes to complex natural products, for which much more far-sighted, multistep planning is necessary15,16 and closely related literature precedents cannot be relied on. Here we demonstrate that such computational synthesis planning is possible, provided that the program's knowledge of organic chemistry and data-based artificial intelligence routines are augmented with causal relationships17,18, allowing it to 'strategize' over multiple synthetic steps. Using a Turing-like test administered to synthesis experts, we show that the routes designed by such a program are largely indistinguishable from those designed by humans. We also successfully validated three computer-designed syntheses of natural products in the laboratory. Taken together, these results indicate that expert-level automated synthetic planning is feasible, pending continued improvements to the reaction knowledge base and further code optimization.

Potential for Standardization and Automation for Pathology and Endoscopy in Inflammatory Bowel Disease.

Automated image analysis methods have shown potential for replicating expert interpretation of histology and endoscopy images, which traditionally require highly specialized and experienced reviewers. Inflammatory bowel disease (IBD) diagnosis, severity assessment, and treatment decision-making require multimodal expert data interpretation and integration, which could be significantly aided by applications of machine learning analyses. This review introduces fundamental concepts of machine learning for imaging analysis and highlights research and development of automated histology and endoscopy interpretation in IBD. Proof-of-concept studies strongly suggest that histologic and endoscopic images can be interpreted with similar accuracy as knowledge experts. Encouraging results support the potential of automating existing disease activity scoring instruments with high reproducibility, speed, and accessibility, therefore improving the standardization of IBD assessment. Though challenges surrounding ground truth definitions, technical barriers, and the need for extensive multicenter evaluation must be resolved before clinical implementation, automated image analysis is likely to both improve access to standardized IBD assessment and advance the fundamental concepts of how disease is measured.

Diagnostic yield and clinical utility of whole exome sequencing using an automated variant prioritization system, EVIDENCE.

EVIDENCE, an automated variant prioritization system, has been developed to facilitate whole exome sequencing analyses. This study investigated the diagnostic yield of EVIDENCE in patients with suspected genetic disorders. DNA from 330 probands (age range, 0-68 years) with suspected genetic disorders were subjected to whole exome sequencing. Candidate variants were identified by EVIDENCE and confirmed by testing family members and/or clinical reassessments. EVIDENCE reported a total 228 variants in 200 (60.6%) of the 330 probands. The average number of organs involved per patient was 4.5 ± 5.0. After clinical reassessment and/or family member testing, 167 variants were identified in 141 probands (42.7%), including 105 novel variants. These variants were confirmed as being responsible for 121 genetic disorders. A total of 103 (61.7%) of the 167 variants in 95 patients were classified as pathogenic or probably to be pathogenic before, and 161 (96.4%) variants in 137 patients (41.5%) after, clinical assessment and/or family member testing. Factor associated with a variant being regarded as causative includes similar symptom scores of a gene variant to the phenotype of the patient. This new, automated variant interpretation system facilitated the diagnosis of various genetic diseases with a 42.7% diagnostic yield.

Safety, time and cost evaluation of automated and semi-automated drug distribution systems in hospitals: a systematic review.

OBJECTIVES: To systematically review automated and semi-automated drug distribution systems (DDSs) in hospitals and to evaluate their effectiveness on medication safety, time and costs of medication care. METHODS: A systematic literature search was conducted in MEDLINE Ovid, Scopus, CINAHL and EMB Reviews covering the period 2005 to May 2016. Studies were included if they (1) concerned technologies used in the drug distribution and administration process in acute care hospitals and (2) reported medication safety, time and cost-related outcomes. RESULTS: Key outcomes, conclusions and recommendations of the included studies (n=30) were categorised according to the dispensing method: decentralised (n=19 studies), centralised (n=6) or hybrid system (n=5). Patient safety improved (n=27) with automation, and reduction in medication errors was found in all three systems. Centralised and decentralised systems were reported to support clinical pharmacy practice in hospitals. The impact of the medication distribution system on time allocation such as labour time, staffing workload or changes in work process was explored in the majority of studies (n=24). Six studies explored economic outcomes. CONCLUSIONS: No medication distribution system was found to be better than another in terms of outcomes assessed in the studies included in the systematic review. All DDSs improved medication safety and quality of care, mainly by decreasing medication errors. However, many error types still remained-for example, prescribing errors. Centralised and hybrid systems saved more time than a decentralised system. Costs of medication care were reduced in decentralised systems mainly in high-expense units. However, no evidence was shown that implementation of decentralised systems in small units would save costs. More comparable evidence on the benefits and costs of decentralised and hybrid systems should be available. Changes in processes due to a new DDS may create new medication safety risks; to minimise these risks, training and reallocation of staff resources are needed.

An integrated architecture for intelligence evaluation of automated vehicles.

Increasing automation calls for evaluating the effectiveness and intelligence of automated vehicles. This paper proposes a framework for quantitatively evaluating the intelligence of automated vehicles. Firstly, we establish the evaluation environment for automated vehicles including test field, test task, and evaluation index. The test tasks include the single vehicle decision-making (turning, lane-changing, overtaking, etc.) and the maneuver execution of multi-vehicle interaction (obstacle avoidance, trajectory optimization, etc.). The intelligence evaluation index is the action amount of driving process considering the safety, efficiency, rationality and comfort. Then, we calculate the actual action amount of the automated vehicle in different scenarios in the test field. Finally, the least action calculated theoretically corresponds to the highest intelligence degree of the automated vehicle, and is employed as a standard to quantify the performance of other tested automated vehicles. The effectiveness of this framework is verified with two naturalistic driving datasets that contain the normal driving scenarios and high-risk scenarios. Specifically, the naturalistic lane-changing data filters 40,416 frames and 179 similar lane-changing trajectories. Compared with the lane-changing behavior of a large number of drivers, experimental results verify that the proposed algorithm can achieve the intelligence degree of drivers in the lane change scenario. Meanwhile, in 253 reconstructed high-risk scenarios, the intelligent risk avoidance ability of the proposed intelligence degree evaluation algorithm can be verified by comparing with the driver behavior and TTC algorithm. These experimental results show that the proposed framework can effectively quantify intelligence and evaluate the performance of automated vehicles under various scenarios.

Automation reliability, human-machine system performance, and operator compliance: A study with airport security screeners supported by automated explosives detection systems for cabin baggage screening.

Using a simulated X-ray screening task, we tested 122 airport security screeners working with the support of explosives detection systems for cabin baggage screening (EDSCB) as low-level automation. EDSCB varied systematically on three automation reliability measures: accuracy, d', and positive predictive value (PPV). Results showed that when unaided performance was high, operator confidence was high, and automation provided only small benefits. When unaided performance was lower, operator confidence was lower, and automation with higher d' provided large benefits. Operator compliance depended on the PPV of automation: We found lower compliance for lower PPV. Automation with a high false alarm rate of 20% and a low PPV of .3 resulted in operators ignoring about one-half of the true automation alarms on difficult targets-a strong cry-wolf effect. Our results suggest that automation reliability described by d' and PPV is more valid than using accuracy alone. When the PPV is below .5, operators should receive clear instructions on how to respond to automation alarms.

Automated thresholding algorithms outperform manual thresholding in macular optical coherence tomography angiography image analysis.

INTRODUCTION: For quantification of Optical Coherence Tomography Angiography (OCTA) images, Vessel Density (VD) and Vessel Skeleton Density (VSD) are well established parameters and different algorithms are in use for their calculation. However, comparability, reliability and ability to discriminate healthy and impaired macular perfusion of different algorithms are unclear, yet, of potential high clinical relevance. Hence, we assessed comparability and test-retest reliability of the most common approaches. MATERIALS AND METHODS: Two consecutive 3×3mm OCTA en face images of the superficial and deep retinal layer were acquired with swept-source OCTA. VD and VSD were calculated with manual thresholding and six automated thresholding algorithms (Huang, Li, Otsu, Moments, Mean, Percentile) using ImageJ and compared in terms of intra-class correlation coefficients, measurement differences and repeatability coefficients. Receiver operating characteristic analyses (healthy vs. macular pathology) were performed and Area Under the Curve (AUC) values were calculated. RESULTS: Twenty-six eyes (8 female, mean age: 47 years) of 15 patients were included (thereof 15 eyes with macular pathology). Binarization thresholds, VD and VSD differed significantly between the algorithms and compared to manual thresholding (p < 0.0001). Inter-measurement differences did not differ significantly between patients with healthy versus pathologic maculae (p ≥ 0.685). Reproducibility was higher for the automated algorithms compared to manual thresholding on all measures of reproducibility assessed. AUC was significantly higher for the Mean algorithm compared to the manual approach with respect to the superficial retinal layer. CONCLUSIONS: Automated thresholding algorithms yield a higher reproducibility of OCTA parameters and allow for a more sensitive diagnosis of macular pathology. However, different algorithms are not interchangeable nor results readily comparable. Especially the Mean algorithm should be investigated in further detail. Automated thresholding algorithms are preferable but more standardization is needed for clinical use.

An open-source drug discovery platform enables ultra-large virtual screens.

On average, an approved drug currently costs US$2-3 billion and takes more than 10 years to develop1. In part, this is due to expensive and time-consuming wet-laboratory experiments, poor initial hit compounds and the high attrition rates in the (pre-)clinical phases. Structure-based virtual screening has the potential to mitigate these problems. With structure-based virtual screening, the quality of the hits improves with the number of compounds screened2. However, despite the fact that large databases of compounds exist, the ability to carry out large-scale structure-based virtual screening on computer clusters in an accessible, efficient and flexible manner has remained difficult. Here we describe VirtualFlow, a highly automated and versatile open-source platform with perfect scaling behaviour that is able to prepare and efficiently screen ultra-large libraries of compounds. VirtualFlow is able to use a variety of the most powerful docking programs. Using VirtualFlow, we prepared one of the largest and freely available ready-to-dock ligand libraries, with more than 1.4 billion commercially available molecules. To demonstrate the power of VirtualFlow, we screened more than 1 billion compounds and identified a set of structurally diverse molecules that bind to KEAP1 with submicromolar affinity. One of the lead inhibitors (iKeap1) engages KEAP1 with nanomolar affinity (dissociation constant (Kd) = 114 nM) and disrupts the interaction between KEAP1 and the transcription factor NRF2. This illustrates the potential of VirtualFlow to access vast regions of the chemical space and identify molecules that bind with high affinity to target proteins.

Optimization of XCMS parameters for LC-MS metabolomics: an assessment of automated versus manual tuning and its effect on the final results.

INTRODUCTION: Several software packages containing diverse algorithms are available for processing Liquid Chromatography-Mass Spectrometry (LC-MS) chromatographic data and within these deconvolution packages different parameters settings can lead to different outcomes. XCMS is the most widely used peak picking and deconvolution software for metabolomics, but the parameter selection can be hard for inexpert users. To solve this issue, the automatic optimization tools such as Isotopologue Parameters Optimization (IPO) can be extremely helpful. OBJECTIVES: To evaluate the suitability of IPO as a tool for XCMS parameters optimization and compare the results with those manually obtained by an exhaustive examination of the LC-MS characteristics and performance. METHODS: Raw HPLC-TOF-MS data from two types of biological samples (liver and plasma) analysed in both positive and negative electrospray ionization modes from three groups of piglets were processed with XCMS using parameters optimized following two different approaches: IPO and Manual. The outcomes were compared to determine the advantages and disadvantages of using each method. RESULTS: IPO processing produced the higher number of repeatable (%RSD < 20) and significant features for all data sets and allowed the different piglet groups to be distinguished. Nevertheless, on multivariate level, similar clustering results were obtained by Principal Component Analysis (PCA) when applied to IPO and manual matrices. CONCLUSION: IPO is a useful optimization tool that helps in choosing the appropriate parameters. It works well on data with a good LC-MS performance but the lack of such adequate data can result in unrealistic parameter settings, which might require further investigation and manual tuning. On the contrary, manual selection criteria requires deeper knowledge on LC-MS, programming language and XCMS parameter interpretation, but allows a better fine-tuning of the parameters, and thus more robust deconvolution.

Driving automation state-of-mind: Using training to instigate rapid mental model development.

The automotive industry is chugging along towards full autonomy, with a yet unknown time of arrival. The next call, however, is partial driving automation. At this interim station lurks many dangers, there-among them issues surrounding the partial performance of the driving task. Despite their potential for increased safety, these systems come with many inherent limitations and caveats, and their safe use depend on drivers correctly understanding their new role. Training is proposed as a potentially effective method of introducing drivers to the central aspects in this human-automation interaction. A proof-of-concept training program designed to introduce drivers to a partial automation system was developed. The effects of training were then evaluated through a between-group mixed-methods simulator experiment. Results indicate that trained drivers both self-report and exhibit an improved understanding of the automation system. They also report a significantly higher inclination to retake control in critical situation, than do their untrained counterparts.

A Super-Clustering Approach for Fully Automated Single Particle Picking in Cryo-EM.

Structure determination of proteins and macromolecular complexes by single-particle cryo-electron microscopy (cryo-EM) is poised to revolutionize structural biology. An early challenging step in the cryo-EM pipeline is the detection and selection of particles from two-dimensional micrographs (particle picking). Most existing particle-picking methods require human intervention to deal with complex (irregular) particle shapes and extremely low signal-to-noise ratio (SNR) in cryo-EM images. Here, we design a fully automated super-clustering approach for single particle picking (SuperCryoEMPicker) in cryo-EM micrographs, which focuses on identifying, detecting, and picking particles of the complex and irregular shapes in micrographs with extremely low signal-to-noise ratio (SNR). Our method first applies advanced image processing procedures to improve the quality of the cryo-EM images. The binary mask image-highlighting protein particles are then generated from each individual cryo-EM image using the super-clustering (SP) method, which improves upon base clustering methods (i.e., k-means, fuzzy c-means (FCM), and intensity-based cluster (IBC) algorithm) via a super-pixel algorithm. SuperCryoEMPicker is tested and evaluated on micrographs of ß-galactosidase and 80S ribosomes, which are examples of cryo-EM data exhibiting complex and irregular particle shapes. The results show that the super-particle clustering method provides a more robust detection of particles than the base clustering methods, such as k-means, FCM, and IBC. SuperCryoEMPicker automatically and effectively identifies very complex particles from cryo-EM images of extremely low SNR. As a fully automated particle detection method, it has the potential to relieve researchers from laborious, manual particle-labeling work and therefore is a useful tool for cryo-EM protein structure determination.

Automation of the I-PASS Tool to Improve Transitions of Care.

INTRODUCTION: Errors in communication during handoffs are a significant source of medical error and put patients at risk. The I-PASS system was designed to systematically communicate information to the oncoming healthcare provider and has been shown to decrease the risk of communication errors. The objective of this observational quality improvement study was to determine whether the addition of a partially automated, electronic handoff tool would further decrease errors in communication during transitions of care for inpatient medical teams. METHODS: We created an electronic tool to incorporate user-generated patient information in the I-PASS format with automatically compiled data derived from the electronic medical record. Numbers of errors in the printed handoff document were recorded before and after intervention. RESULTS: The first implementation cycle demonstrated an absolute risk reduction for written errors of 45.6% (95% confidence interval [CI] 39.2-51.2%) and a number needed to treat (NNT) of three patients. The second cycle showed an absolute risk reduction of 53.3% (95% CI 39.8-63.9%; NNT 2). Aggregate data showed an absolute risk reduction of 46.6% (95% CI 41.0-51.7%, NNT 3). CONCLUSIONS: Improving the routine task of patient handoff through the thoughtful application of technology can yield benefits in terms of decreasing documentation errors and streamlining workflow before patient handoff.

Toward systematic review automation: a practical guide to using machine learning tools in research synthesis.

Technologies and methods to speed up the production of systematic reviews by reducing the manual labour involved have recently emerged. Automation has been proposed or used to expedite most steps of the systematic review process, including search, screening, and data extraction. However, how these technologies work in practice and when (and when not) to use them is often not clear to practitioners. In this practical guide, we provide an overview of current machine learning methods that have been proposed to expedite evidence synthesis. We also offer guidance on which of these are ready for use, their strengths and weaknesses, and how a systematic review team might go about using them in practice.

Towards Automated Manufacturing for Cell Therapies.

PURPOSE OF REVIEW: Many cell therapy products are beginning to reach the commercial finish line and a rapidly escalating pipeline of products are in clinical development. The need to develop manufacturing capability that will support a successful commercial business model has become a top priority as many cell therapy developers look to secure long-term visions to enable both funding and treatment success. RECENT FINDINGS: Manufacturing automation is both highly compelling and very challenging at the same time as a key tactic to address quality, cost of goods, scale, and sustainability that are fundamental drivers for commercially viable manufacturing. This paper presents an overview and strategic drivers for application of automation to cell therapy manufacturing. It also explores unique automation considerations for patient-specific cell therapy (PSCT) where each full-scale lot is for one patient vs off-the-shelf cell therapy (OTSCT) where a full-scale lot will treat many patients, and finally some practical considerations for implementing automation.

Automated EEG source imaging: A retrospective, blinded clinical validation study.

OBJECTIVE: To evaluate the accuracy of automated EEG source imaging (ESI) in localizing epileptogenic zone. METHODS: Long-term EEG, recorded with the standard 25-electrode array of the IFCN, from 41 consecutive patients with focal epilepsy who underwent resective surgery, were analyzed blinded to the surgical outcome. The automated analysis comprised spike-detection, clustering and source imaging at the half-rising time and at the peak of each spike-cluster, using individual head-models with six tissue-layers and a distributed source model (sLORETA). The fully automated approach presented ESI of the cluster with the highest number of spikes, at the half-rising time. In addition, a physician involved in the presurgical evaluation of the patients, evaluated the automated ESI results (up to four clusters per patient) in clinical context and selected the dominant cluster and the analysis time-point (semi-automated approach). The reference standard was location of the resected area and outcome one year after operation. RESULTS: Accuracy was 61% (95% CI: 45-76%) for the fully automated approach and 78% (95% CI: 62-89%) for the semi-automated approach. CONCLUSION: Automated ESI has an accuracy similar to previously reported neuroimaging methods. SIGNIFICANCE: Automated ESI will contribute to increased utilization of source imaging in the presurgical evaluation of patients with epilepsy.