AxonDeep: Automated Optic Nerve Axon Segmentation in Mice with Deep Learning.

Purpose: Optic nerve damage is the principal feature of glaucoma and contributes to vision loss in many diseases. In animal models, nerve health has traditionally been assessed by human experts that grade damage qualitatively or manually quantify axons from sampling limited areas from histologic cross sections of nerve. Both approaches are prone to variability and time consuming. Automated approaches have begun to emerge, but shortcomings have limited wide-spread application. Here, we seek improvements through use of deep-learning approaches for segmenting and quantifying axons from cross sections of mouse optic nerve. Methods: Two deep-learning approaches were developed and evaluated: (1) a traditional supervised approach using a fully convolutional network trained with only labeled data and (2) a semi-supervised approach trained with both labeled and unlabeled data using a generative-adversarial-network framework. Results: From comparisons with an independent test set of images with manually marked axon centers and boundaries, both deep-learning approaches performed above an existing baseline automated approach and similarly to two independent experts. Performance of the semi-supervised approach was superior and implemented into AxonDeep. Conclusions: AxonDeep performs automated quantification and segmentation of axons similar to that of experts without the time- and labor-constraints associated with manual performance. The quantitative and objective nature of AxonDeep reduces variability arising from differences in model, methodology, and user that often compromise manual performance of these tasks. Translational Relevance: Use of deep learning for axon quantification provides rapid, objective, and higher throughput analysis of optic nerve that would otherwise not be possible.

Should We Just Let the Machines Do It? The Benefit and Cost of Action Recommendation and Action Implementation Automation

Objective To examine the effects of action recommendation and action implementation automation on performance, workload, situation awareness (SA), detection of automation failure, and return-to-manual performance in a submarine track management task. Background Theory and meta-analytic evidence suggest that with increasing degrees of automation (DOA), operator performance improves and workload decreases, but SA and return-to-manual performance declines. Method Participants monitored the location and heading of contacts in order to classify them, mark their closest point of approach (CPA), and dive when necessary. Participants were assigned either no automation, action recommendation automation, or action implementation automation. An automation failure occurred late in the task, whereby the automation provided incorrect classification advice or implemented incorrect classification actions. Results Compared to no automation, action recommendation automation benefited automated task performance and lowered workload, but cost nonautomated task performance. Action implementation automation resulted in perfect automated task performance (by default) and lowered workload, with no costs to nonautomated task performance, SA, or return-to-manual performance compared to no automation. However, participants provided action implementation automation were less likely to detect the automation failure compared to those provided action recommendations, and made less accurate classifications immediately after the automation failure, compared to those provided no automation. Conclusion Action implementation automation produced the anticipated benefits but also caused poorer automation failure detection. Application While action implementation automation may be effective for some task contexts, system designers should be aware that operators may be less likely to detect automation failures and that performance may suffer until such failures are detected.

Effects of Scaling and Sequence on Performance of Dynamic Bimanual Path Following Tasks

Bimanual coordination plays a vital role in many haptic and robotic system operations. However, theories in bimanual human motor control are rarely integrated into the control system for human-in-the-loop robots, potentially limiting the usability and collaborative potential between the human and robot, particularly for complex tasks such as robotic surgery. To inform future integration, we investigate unknown manual performance relationships regarding the scaling (the size of one hand’s motions compared to the other) and sequence (the order in which the hands move) of complex bimanual path following tasks. For scaling variations, either the left or right hand desired trajectory amplitude was increased. For sequence, the task was split so that the hands moved sequentially or simultaneously. The experiment is performed by 11 inexperienced, able bodied subjects (all right-handed) in a virtual environment while using haptic devices. Results show significant ([Formula: see text]) decreased manual performance for one hand when the opposite hand is scaled, thus suggesting an increase in the scale of one hand will decrease the performance of the contralateral. Results also show a significant decrease in performance for the left hand when moving simultaneous with the right, but the right hand does not show such a decrease in performance. This might suggest that only the nondominant hand suffers from simultaneous motion conditions. These results may lead to unique opportunities to integrate theories related to human motor control into the control system for haptic and robotic systems used in complex bimanual upper-limb tasks.

Reliability of integrated parameters of operational efficiency of TPP power units

A new method and an algorithm of ensuring the uniformity of normalized samples of technical and economic parameters of power units of thermal power plants are presented. Uniformity and normalization are obligatory conditions at estimation of integrated parameters characterizing the efficiency of power units. The method is based on the fiducial approach. Boundary values of fiducial interval are traditionally calculated based on the statistical function of distribution and a set significance value, i.e. are calculated, in fact, "mechanically". As the "mechanical" approach is appropriate for homogeneous statistical data, whereas technical and economic parameters are multivariate data, application of this approach to the statistical function of fiducial distribution is associated with a high risk of an erroneous decision. The set of possible realizations of actual values of technical and economic parameters includes realizations caused by "gross" mistakes at data input in automated systems or at manual performance of individual calculations. Quite often, unconventional realizations are observed, e. g., in case of low-load operation for 10 days of a month. These data form boundary intervals that the authors refer to as “boundins” (or prigrins in Russian). Automated search and removal of boundins provides reliability of comparison and ranking of integrated parameters. It is shown that rate of variation of boundins is considerably below the rate of variation of typical realizations of technical and economic parameters. This fact was the basis for recognition of boundins.

Intrathecal Infusion of Autologous Adipose-Derived Regenerative Cells in Autoimmune Refractory Epilepsy: Evaluation of Safety and Efficacy

Objective/Purpose. Evaluation of efficacy and safety of autologous adipose-derived regenerative cells (ADRCs) treatment in autoimmune refractory epilepsy. Patients. Six patients with proven or probable autoimmune refractory epilepsy (2 with Rasmussen encephalitis, 2 with antineuronal autoantibodies in serum, and 2 with possible FIRES) were included in the project with approval of the Bioethics Committee. Method. Intrathecal injection of autologous ADRC acquired through liposuction followed by enzymatic isolation was performed. The procedure was repeated 3 times every 3 months with each patient. Neurological status, brain MRI, cognitive function, and antiepileptic effect were monitored during 12 months. Results. Immediately after the procedure, all patients were in good condition. In some cases, transient mildly elevated body temperature, pain in regions of liposuction, and slight increasing number of seizures during 24 hours were observed. During the next months, some improvements in school, social functioning, and manual performance were observed in all patients. One patient has been seizure free up to the end of trial. In other patients, frequency of seizures was different: from reduced number to the lack of improvement (3-year follow-up). Conclusion. Autologous ADRC therapy may emerge as a promising option for some patients with autoimmune refractory epilepsy. Based on our trial and other clinical data, the therapy appears to be safe and feasible. Antiepileptic efficacy proved to be various; however, some abilities improved in all children. No signs of psychomotor regression were observed during the first year following the treatment.

The Benefits and Costs of Low and High Degree of Automation

Objective The objective of this study is to examine the effects of low and high degree of automation (DOA) on performance, subjective workload, situation awareness (SA), and return-to-manual control in simulated submarine track management. Background Theory and meta-analytic evidence suggest that as DOA increases, operator performance improves and workload decreases, but SA and return-to-manual control declines. Research also suggests that operators have particular difficulty regaining manual control if automation provides incorrect advice. Method Undergraduate student participants completed a submarine track management task that required them to track the position and behavior of contacts. Low DOA supported information acquisition and analysis, whereas high DOA recommended decisions. At a late stage in the task, automation was either unexpectedly removed or provided incorrect advice. Results Relative to no automation, low DOA moderately benefited performance but impaired SA and non-automated task performance. Relative to no automation and low DOA, high DOA benefited performance and lowered workload. High DOA did impair non-automated task performance compared with no automation, but this was equivalent to low DOA. Participants were able to return-to-manual control when they knew low or high DOA was disengaged, or when high DOA provided incorrect advice. Conclusion High DOA improved performance and lowered workload, at no additional cost to SA or return-to-manual performance when compared with low DOA. Application Designers should consider the likely level of uncertainty in the environment and the consequences of return-to-manual deficits before implementing low or high DOA.

The Effect of the Degree of Static Automation in Simulated Submarine Track Management

Automation that supports our workplaces is intended to relieve the requirement for humans to control tasks, as a way to reduce operator workload and maximize system capacity. Researchers have long recognized the potential costs associated with automation. These costs include the loss of an operator’s understanding of a task and an inability to anticipate future task events ( situation awareness; SA; Endsley, 1995) that can occur due to automation induced complacency (Parasuraman, Molloy, & Singh, 1993), and the subsequent lack of ability to regain manual control after automation (Kaber & Endsley, 2004). These costs to automation are more likely to occur when the degree of automation (DOA) increases. DOA has been defined based on whether automation is doing more or less ‘work’ ( levels of automation; Sheridan & Verplank, 1978), and at which of the four stages of human information processing the automation is directed; information acquisition, information analysis, decision selection, and action implementation ( stages of automation; Parasuraman, Sheridan, & Wickens, 2000). As the DOA increases, performance and workload tend to improve. However, SA and return-to-manual performance can decline. Recent research by Chen, Huf, Visser, and Loft (2017) reported that a low DOA had minimal benefits to performance and workload, and also impaired SA and non-automated task performance compared to a manual control condition in a simulated submarine track management task. However, the low DOA did not lead to any return-to-manual deficits when automation was unexpectedly removed. The current study compared the effects of low and high DOA on operator performance, workload, SA, non-automated task performance, and return-to-manual performance in submarine track management. Participants ( N= 122) monitored a tactical display that presented the location and heading of contacts in relation to the Ownship and landmarks, and a ‘waterfall’ display that presented sonar bearings of contacts and how those bearings change with time. Participants performed three tasks: classification, closest point of approach (CPA), and dive. The classification task involved classifying contacts depending on how long they had spent within display regions. The CPA task involved monitoring changes in contact heading to determine their closest point of approach to the Ownship. The dive task involved integrating contact location and heading information to determine when the submarine could safely dive. Automated assistance was provided for the classification and CPA tasks, but not for the dive task. The low DOA condition received information acquisition and analysis support (stages 1 and 2), whereas the high DOA received decision selection support (stage 3). In a mixed design, the between-subjects factor was condition (no automation, high DOA, low DOA) and the within-subjects factor was automation state (routine, automation removal). Participants completed three track management scenarios, and during the last scenario the automation was unexpectedly removed. Firstly, we predicted that a high DOA would have larger benefits to performance and workload compared to a low DOA, but that these benefits might be accompanied by costs to SA, non-automated task performance, and return-to-manual performance. Secondly, we predicted that a low DOA would show minimal benefits to performance and workload, significant costs to SA and non-automated task performance, and no effect on return-to-manual performance when compared to no automation, thus replicating the findings of Chen et al. (2017). The results from this study indicated that relative to the low DOA condition, participants provided with high DOA support had better performance and lower workload, without any further costs to SA, non-automated task performance, or return-to-manual performance. Furthermore, relative to no automation, participants provided with low DOA support only had minor benefits to performance (replicating Chen et al., 2017) and no benefits to workload, and significant costs to SA and non-automated task performance. In summary, the high DOA produced larger benefits to performance and workload than the low DOA, without increasing costs. In light of these results, the automated system that recommended decisions was effectively utilized by operators in the current context, and appeared to be superior to the automated system that supported information acquisition and analysis.