Applying Machine Learning to the Visual Inspection of Filled Injectable Drug Products.

With machine learning (ML), we see the potential to better harness the intelligence and decision-making abilities of human inspectors performing manual visual inspection (MVI) and apply this to automated visual inspection (AVI) with the inherent improvements in throughput and consistency. This article is intended to capture current experience with this new technology and provides points to consider for successful application to AVI of injectable drug products. The technology is available today for such AVI applications. Machine vision companies have integrated ML as an additional visual inspection tool with minimal upgrades to existing hardware. Studies have demonstrated superior results in defect detection and reduction in false rejects, when compared with conventional inspection tools. ML implementation does not require modifications to current AVI qualification strategies. The utilization of this technology for AVI will accelerate recipe development by use of faster computers rather than by direct human configuration and coding of vision tools. By freezing the model developed with artificial intelligence tools and subjecting it to current validation strategies, assurance of reliable performance in the production environment can be achieved.

About the cortical origin of the low-delta and high-gamma rhythms observed in EEG signals during treadmill walking.

This paper presents a spectral and time-frequency analysis of EEG signals recorded on seven healthy subjects walking on a treadmill at three different speeds. An accelerometer was placed on the head of the subjects in order to record the shocks undergone by the EEG electrodes during walking. Our results indicate that up to 15 harmonics of the fundamental stepping frequency may pollute EEG signals, depending on the walking speed and also on the electrode location. This finding may call into question some conclusions drawn in previous EEG studies where low-delta band (especially around 1 Hz, the fundamental stepping frequency) had been announced as being the seat of angular and linear kinematics control of the lower limbs during walk. Additionally, our analysis reveals that EEG and accelerometer signals exhibit similar time-frequency properties, especially in frequency bands extending up to 150 Hz, suggesting that previous conclusions claiming the activation of high-gamma rhythms during walking may have been drawn on the basis of insufficiently cleaned EEG signals. Our results are put in perspective with recent EEG studies related to locomotion and extensively discussed in particular by focusing on the low-delta and high-gamma bands.

Performance of the Emotiv Epoc headset for P300-based applications.

BACKGROUND: For two decades, EEG-based Brain-Computer Interface (BCI) systems have been widely studied in research labs. Now, researchers want to consider out-of-the-lab applications and make this technology available to everybody. However, medical-grade EEG recording devices are still much too expensive for end-users, especially disabled people. Therefore, several low-cost alternatives have appeared on the market. The Emotiv Epoc headset is one of them. Although some previous work showed this device could suit the customer's needs in terms of performance, no quantitative classification-based assessments compared to a medical system are available. METHODS: This paper aims at statistically comparing a medical-grade system, the ANT device, and the Emotiv Epoc headset by determining their respective performances in a P300 BCI using the same electrodes. On top of that, a review of previous Emotiv studies and a discussion on practical considerations regarding both systems are proposed. Nine healthy subjects participated in this experiment during which the ANT and the Emotiv systems are used in two different conditions: sitting on a chair and walking on a treadmill at constant speed. RESULTS: The Emotiv headset performs significantly worse than the medical device; observed effect sizes vary from medium to large. The Emotiv headset has higher relative operational and maintenance costs than its medical-grade competitor. CONCLUSIONS: Although this low-cost headset is able to record EEG data in a satisfying manner, it should only be chosen for non critical applications such as games, communication systems, etc. For rehabilitation or prosthesis control, this lack of reliability may lead to serious consequences. For research purposes, the medical system should be chosen except if a lot of trials are available or when the Signal-to-Noise Ratio is high. This also suggests that the design of a specific low-cost EEG recording system for critical applications and research is still required.

Biological oscillations for learning walking coordination: dynamic recurrent neural network functionally models physiological central pattern generator.

The existence of dedicated neuronal modules such as those organized in the cerebral cortex, thalamus, basal ganglia, cerebellum, or spinal cord raises the question of how these functional modules are coordinated for appropriate motor behavior. Study of human locomotion offers an interesting field for addressing this central question. The coordination of the elevation of the 3 leg segments under a planar covariation rule (Borghese et al., 1996) was recently modeled (Barliya et al., 2009) by phase-adjusted simple oscillators shedding new light on the understanding of the central pattern generator (CPG) processing relevant oscillation signals. We describe the use of a dynamic recurrent neural network (DRNN) mimicking the natural oscillatory behavior of human locomotion for reproducing the planar covariation rule in both legs at different walking speeds. Neural network learning was based on sinusoid signals integrating frequency and amplitude features of the first three harmonics of the sagittal elevation angles of the thigh, shank, and foot of each lower limb. We verified the biological plausibility of the neural networks. Best results were obtained with oscillations extracted from the first three harmonics in comparison to oscillations outside the harmonic frequency peaks. Physiological replication steadily increased with the number of neuronal units from 1 to 80, where similarity index reached 0.99. Analysis of synaptic weighting showed that the proportion of inhibitory connections consistently increased with the number of neuronal units in the DRNN. This emerging property in the artificial neural networks resonates with recent advances in neurophysiology of inhibitory neurons that are involved in central nervous system oscillatory activities. The main message of this study is that this type of DRNN may offer a useful model of physiological central pattern generator for gaining insights in basic research and developing clinical applications.

Towards effective non-invasive brain-computer interfaces dedicated to gait rehabilitation systems.

In the last few years, significant progress has been made in the field of walk rehabilitation. Motor cortex signals in bipedal monkeys have been interpreted to predict walk kinematics. Epidural electrical stimulation in rats and in one young paraplegic has been realized to partially restore motor control after spinal cord injury. However, these experimental trials are far from being applicable to all patients suffering from motor impairments. Therefore, it is thought that more simple rehabilitation systems are desirable in the meanwhile. The goal of this review is to describe and summarize the progress made in the development of non-invasive brain-computer interfaces dedicated to motor rehabilitation systems. In the first part, the main principles of human locomotion control are presented. The paper then focuses on the mechanisms of supra-spinal centers active during gait, including results from electroencephalography, functional brain imaging technologies [near-infrared spectroscopy (NIRS), functional magnetic resonance imaging (fMRI), positron-emission tomography (PET), single-photon emission-computed tomography (SPECT)] and invasive studies. The first brain-computer interface (BCI) applications to gait rehabilitation are then presented, with a discussion about the different strategies developed in the field. The challenges to raise for future systems are identified and discussed. Finally, we present some proposals to address these challenges, in order to contribute to the improvement of BCI for gait rehabilitation.

A subjective assessment of a P300 BCI system for lower-limb rehabilitation purposes.

Recent research has shown that a P300 system can be used while walking without requiring any specific gait-related artifact removal techniques. Also, standard EEG-based Brain-Computer Interfaces (BCI) have not been really assessed for lower limb rehabilitation/prosthesis. Therefore, this paper gives a first baseline estimation (for future BCI comparisons) of the subjective and objective performances of a four-state P300 BCI plus a non-control state for lower-limb rehabilitation purposes. To assess usability and workload, the System Usability Scale and the NASA Task Load Index questionnaires were administered to five healthy subjects after performing a real-time treadmill speed control. Results show that the P300 BCI approach could suit fitness and rehabilitation applications, whereas prosthesis control, which suffers from a low reactivity, appears too sensitive for risky and crowded areas.

An active foot lifter orthosis based on a PCPG algorithm.

Central pattern generators (CPGs) are known to play an important role in the generation of rhythmic movements in gait, both in animals and humans. The comprehension of their underlying mechanism has led to the development of an important family of algorithms at the basis of autonomous walking robots. Recently, it has been shown that human gait could be modeled using a subclass of those algorithms, namely a Programmable Central Pattern Generator (PCPG). In this paper, we present a foot lifter orthosis driven by this algorithm. After a learning phase, the PCPG is able to generate adequate rhythmic gait patterns both for constant speeds and acceleration phases. Its output is used to drive the orthosis actuator during the swing phase, in order to help patients suffering from foot drop (the orthosis just follows the movement during the stance phase). The most interesting property of this algorithm is the possibility to generate a smooth output signal even during speed transitions. In practice, given that human gait is not perfectly periodic, the phase of this signal needs to be reset with actual movement. Therefore, two phase-resetting procedures were studied: one standard hard phase-resetting leading to discontinuities and one original soft phase-resetting allowing to recover the correct phase in a smooth way. The simulation results and complete design of the orthosis hardware and software are presented.