Machine learning seizure prediction: one problematic but accepted practice.

Objective.Epilepsy is one of the most common neurological disorders and can have a devastating effect on a person's quality of life. As such, the search for markers which indicate an upcoming seizure is a critically important area of research which would allow either on-demand treatment or early warning for people suffering with these disorders. There is a growing body of work which uses machine learning methods to detect pre-seizure biomarkers from electroencephalography (EEG), however the high prediction rates published do not translate into the clinical setting. Our objective is to investigate a potential reason for this.Approach.We conduct an empirical study of a commonly used data labelling method for EEG seizure prediction which relies on labelling small windows of EEG data in temporal groups then selecting randomly from those windows to validate results. We investigate a confound for this approach for seizure prediction and demonstrate the ease at which it can be inadvertently learned by a machine learning system.Main results.We find that non-seizure signals can create decision surfaces for machine learning approaches which can result in false high prediction accuracy on validation datasets. We prove this by training an artificial neural network to learn fake seizures (fully decoupled from biology) in real EEG.Significance.The significance of our findings is that many existing works may be reporting results based on this confound and that future work should adhere to stricter requirements in mitigating this confound. The problematic, but commonly accepted approach in the literature for seizure prediction labelling is potentially preventing real advances in developing solutions for these sufferers. By adhering to the guidelines in this paper future work in machine learning seizure prediction is more likely to be clinically relevant.

White matter tract conductivity is resistant to wide variations in paranodal structure and myelin thickness accompanying the loss of Tyro3: an experimental and simulated analysis.

Myelination within the central nervous system (CNS) is crucial for the conduction of action potentials by neurons. Variation in compact myelin morphology and the structure of the paranode are hypothesised to have significant impact on the speed of action potentials. There are, however, limited experimental data investigating the impact of changes in myelin structure upon conductivity in the central nervous system. We have used a genetic model in which myelin thickness is reduced to investigate the effect of myelin alterations upon action potential velocity. A detailed examination of the myelin ultrastructure of mice in which the receptor tyrosine kinase Tyro3 has been deleted showed that, in addition to thinner myelin, these mice have significantly disrupted paranodes. Despite these alterations to myelin and paranodal structure, we did not identify a reduction in conductivity in either the corpus callosum or the optic nerve. Exploration of these results using a mathematical model of neuronal conductivity predicts that the absence of Tyro3 would lead to reduced conductivity in single fibres, but would not affect the compound action potential of multiple myelinated neurons as seen in neuronal tracts. Our data highlight the importance of experimental assessment of conductivity and suggests that simple assessment of structural changes to myelin is a poor predictor of neural functional outcomes.

Computational Neural Modeling of Auditory Cortical Receptive Fields.

Previous studies have shown that the auditory cortex can enhance the perception of behaviorally important sounds in the presence of background noise, but the mechanisms by which it does this are not yet elucidated. Rapid plasticity of spectrotemporal receptive fields (STRFs) in the primary (A1) cortical neurons is observed during behavioral tasks that require discrimination of particular sounds. This rapid task-related change is believed to be one of the processing strategies utilized by the auditory cortex to selectively attend to one stream of sound in the presence of mixed sounds. However, the mechanism by which the brain evokes this rapid plasticity in the auditory cortex remains unclear. This paper uses a neural network model to investigate how synaptic transmission within the cortical neuron network can change the receptive fields of individual neurons. A sound signal was used as input to a model of the cochlea and auditory periphery, which activated or inhibited integrate-and-fire neuron models to represent networks in the primary auditory cortex. Each neuron in the network was tuned to a different frequency. All neurons were interconnected with excitatory or inhibitory synapses of varying strengths. Action potentials in one of the model neurons were used to calculate the receptive field using reverse correlation. The results were directly compared to previously recorded electrophysiological data from ferrets performing behavioral tasks that require discrimination of particular sounds. The neural network model could reproduce complex STRFs observed experimentally through optimizing the synaptic weights in the model. The model predicts that altering synaptic drive between cortical neurons and/or bottom-up synaptic drive from the cochlear model to the cortical neurons can account for rapid task-related changes observed experimentally in A1 neurons. By identifying changes in the synaptic drive during behavioral tasks, the model provides insights into the neural mechanisms utilized by the auditory cortex to enhance the perception of behaviorally salient sounds.

A detailed, conductance-based computer model of intrinsic sensory neurons of the gastrointestinal tract.

Intrinsic sensory neurons (ISNs) of the enteric nervous system respond to stimuli such as muscle tension, muscle length, distortion of the mucosa, and the chemical content in the lumen. ISNs form recurrent networks that probably drive many intestinal motor patterns and reflexes. ISNs express a large number of voltage- and calcium-gated ion channels, some of which are modified by inflammation or repeated physiological stimuli, but how interactions between different ionic currents in ISNs produce both normal and pathological behaviors in the intestine remains unclear. We constructed a model of ISNs including voltage-gated sodium and potassium channels, N-type calcium channels, big conductance calcium-dependent potassium (BK) channels, calcium-dependent nonspecific cation channels (NSCa), intermediate conductance calcium-dependent potassium (IK) channels, hyperpolarization-activated cation (Ih) channels, and internal calcium dynamics. The model was based on data from the literature and our electrophysiological studies. The model reproduced responses to short or long depolarizing current pulses and responses to long hyperpolarizing current pulses. Sensitivity analysis showed that Ih, IK, NSCa, and BK have the largest influence on the number of action potentials observed during prolonged depolarizations. The model also predicts that changes to the voltage of activation for Ih have a large influence on excitability, but changes to the time constant of activation for Ih have a minor effect. Our model identifies how interactions between different iconic currents influence the excitability of ISNs and highlights an important role for Ih in enteric neuroplasticity resulting from disease.

Mathematical modelling of enteric neural motor patterns.

1. The enteric nervous system modulates intestinal behaviours, such as motor patterns and secretion. Although much is known about different types of neurons and simple reflexes in the intestine, it remains unclear how complex behaviours are generated. 2. Mathematical modelling is an important tool for assisting the understanding of how the neurons and reflexes can be pieced together to generate intestinal behaviours. 3. Models have identified a functional role for slow excitatory post-synaptic potentials (EPSPs) by distinguishing between fast and slow EPSPs in the ascending excitation reflex. These models also discovered coordinated firing of similarly located neurons as emergent properties of feed-forward networks of interneurons in the intestine. A model of the recurrent network of intrinsic sensory neurons identified important control mechanisms to prevent uncontrolled firing due to positive feedback and that the interaction between these control mechanisms and slow EPSPs is necessary for the networks to encode ongoing sensory stimuli. This model also showed that such networks may mediate migrating motor complexes. 4. A network model of vasoactive intestinal peptide neurons in the submucosal plexus found this relatively sparse recurrent network could produce uncontrolled firing under conditions that appear to be related to cholera toxin-induced hypersecretion. 5. Abstract modelling of the intestinal fed-state motor patterns has identified how stationary contractions can arise from a polarized network. 6. These models have also helped predict and/or explained pharmacological evidence for two rhythm generators and the requirement of feedback from contractions in the circular muscle.

Serotonin and cholecystokinin mediate nutrient-induced segmentation in guinea pig small intestine.

Segmentation is an important process in nutrient mixing and absorption; however, the mechanisms underlying this motility pattern are poorly understood. Segmentation can be induced by luminal perfusion of fatty acid in guinea pig small intestine in vitro and mimicked by the serotonin (5-HT) reuptake inhibitor fluoxetine (300 nM) and by cholecystokinin (CCK). Serotonergic and CCK-related mechanisms underlying nutrient-induced segmentation were investigated using selective 5-HT and CCK receptor antagonists on isolated segments of small intestine luminally perfused with 1 mM decanoic acid. Motility patterns were analyzed using video imaging and spatiotemporal maps. Segmenting activity mediated by decanoic acid was depressed following luminal application of the 5-HT receptor antagonists granisetron (5-HT(3), 1 µM) and SB-207266 (5-HT(4), 10 nM) and the CCK receptor antagonists devazepide (CCK-1, 300 nM) and L-365260 (CCK-2, 300 nM), but these antagonists did not further depress segmentation when combined. The P2 receptor antagonist pyridoxal phosphate-6-azophenyl-2',4'-disulfonate (10 µM) had no effect on activity. Serosal application of 5-HT antagonists had little effect on segmentation in the duodenum but reduced activity in the jejunum when granisetron and SB-207266 were applied together. These results reveal that 5-HT(3) and 5-HT(4) receptors, as well as CCK-1 and CCK-2 receptors, are critical in regulating decanoic acid-induced segmentation. Computational simulation indicated that these data are consistent with decanoic acid activating two pathways in the mucosa that converge within the enteric neural circuitry, while contraction-induced release of 5-HT from the mucosa provides feedback into the neural circuit to set the time course of the overall contractile activity.

Parametric computation predicts a multiplicative interaction between synaptic strength parameters that control gamma oscillations.

Gamma oscillations are thought to be critical for a number of behavioral functions, they occur in many regions of the brain and through a variety of mechanisms. Fast repetitive bursting (FRB) neurons in layer 2 of the cortex are able to drive gamma oscillations over long periods of time. Even though the oscillation is driven by FRB neurons, strong feedback within the rest of the cortex must modulate properties of the oscillation such as frequency and power. We used a highly detailed model of the cortex to determine how a cohort of 33 parameters controlling synaptic drive might modulate gamma oscillation properties. We were interested in determining not just the effects of parameters individually, but we also wanted to reveal interactions between parameters beyond additive effects. To prevent a combinatorial explosion in parameter combinations that might need to be simulated, we used a fractional factorial design (FFD) that estimated the effects of individual parameters and two parameter interactions. This experiment required only 4096 model runs. We found that the largest effects on both gamma power and frequency came from a complex interaction between efficacy of synaptic connections from layer 2 inhibitory neurons to layer 2 excitatory neurons and the parameter for the reciprocal connection. As well as the effect of the individual parameters determining synaptic efficacy, there was an interaction between these parameters beyond the additive effects of the parameters alone. The magnitude of this effect was similar to that of the individual parameters, predicting that it is physiologically important in setting gamma oscillation properties.

Multiple neural oscillators and muscle feedback are required for the intestinal fed state motor program.

After a meal, the gastrointestinal tract exhibits a set of behaviours known as the fed state. A major feature of the fed state is a little understood motor pattern known as segmentation, which is essential for digestion and nutrient absorption. Segmentation manifests as rhythmic local constrictions that do not propagate along the intestine. In guinea-pig jejunum in vitro segmentation constrictions occur in short bursts together with other motor patterns in episodes of activity lasting 40-60 s and separated by quiescent episodes lasting 40-200 s. This activity is induced by luminal nutrients and abolished by blocking activity in the enteric nervous system (ENS). We investigated the enteric circuits that regulate segmentation focusing on a central feature of the ENS: a recurrent excitatory network of intrinsic sensory neurons (ISNs) which are characterized by prolonged after-hyperpolarizing potentials (AHPs) following their action potentials. We first examined the effects of depressing AHPs with blockers of the underlying channels (TRAM-34 and clotrimazole) on motor patterns induced in guinea-pig jejunum, in vitro, by luminal decanoic acid. Contractile episode durations increased markedly, but the frequency and number of constrictions within segmenting bursts and quiescent period durations were unaffected. We used these observations to develop a computational model of activity in ISNs, excitatory and inhibitory motor neurons and the muscle. The model predicted that: i) feedback to ISNs from contractions in the circular muscle is required to produce alternating activity and quiescence with the right durations; ii) transmission from ISNs to excitatory motor neurons is via fast excitatory synaptic potentials (EPSPs) and to inhibitory motor neurons via slow EPSPs. We conclude that two rhythm generators regulate segmentation: one drives contractions within segmentation bursts, the other the occurrence of bursts. The latter depends on AHPs in ISNs and feedback to these neurons from contraction of the circular muscle.

Insights into mechanisms of intestinal segmentation in guinea pigs: a combined computational modeling and in vitro study.

Segmentation in the guinea pig small intestine consists of a number of discrete motor patterns including rhythmic stationary contractions that occur episodically at specific locations along the intestine. The enteric nervous system regulates segmentation, but the exact circuit is unknown. Using simple computer models, we investigated possible circuits. Our computational model simulated the mean neuron firing rate in the feedforward ascending and descending reflex pathways. A stimulus-evoked pacemaker was located in the afferent pathway or in a feedforward pathway. Output of the feedforward pathways was fed into a simple model to determine the response of the muscle. Predictions were verified in vitro by using guinea pig jejunum, in which segmentation was induced with luminal fatty acid. In the computational model, local stimuli produced an oral contraction and anal dilation, similar to in vitro responses to local distension, but did not produce segmentation. When the stimulus was distributed, representing a nutrient load, the result was either a tonic response or globally synchronized oscillations. However, when we introduced local variations in synaptic coupling, stationary contractions occurred around these locations. This predicts that severing the ascending and descending pathways will induce stationary contractions. An acute lesion in our in vitro model significantly increased the number of stationary contractions immediately oral and anal to the lesion. Our results suggest that spatially localized rhythmic contractions arise from a local imbalance between ascending excitatory and descending inhibitory muscle inputs and require a distributed stimulus and a rhythm generator in the afferent pathway.

Recurrent networks of submucous neurons controlling intestinal secretion: a modeling study.

Secretomotor neurons, immunoreactive for vasoactive intestinal peptide (VIP), are important in controlling chloride secretion in the small intestine. These neurons form functional synapses with other submucosal VIP neurons and transmit via slow excitatory postsynaptic potentials (EPSPs). Thus they form a recurrent network with positive feedback. Intrinsic sensory neurons within the submucosa are also likely to form recurrent networks with positive feedback, provide substantial output to VIP neurons, and receive input from VIP neurons. If positive feedback within recurrent networks is sufficiently large, then neurons in the network respond to even small stimuli by firing at their maximum possible rate, even after the stimulus is removed. However, it is not clear whether such a mechanism operates within the recurrent networks of submucous neurons. We investigated this question by performing computer simulations of realistic models of VIP and intrinsic sensory neuron networks. In the expected range of electrophysiological properties, we found that activity in the VIP neuron network decayed slowly after cessation of a stimulus, indicating that positive feedback is not strong enough to support the uncontrolled firing state. The addition of intrinsic sensory neurons produced a low stable firing rate consistent with the common finding that basal secretory activity is, in part, neurogenic. Changing electrophysiological properties enables these recurrent networks to support the uncontrolled firing state, which may have implications with hypersecretion in the presence of enterotoxins such as cholera-toxin.