Diagnosis and Severity Assessment of COPD Using a Novel Fast-Response Capnometer and Interpretable Machine Learning

INTRODUCTION

Spirometry is the gold standard for COPD diagnosis and severity determination, but is technique-dependent, nonspecific, and requires administration by a trained healthcare professional. There is a need for a fast, reliable, and precise alternative diagnostic test. This study's aim was to use interpretable machine learning to diagnose COPD and assess severity using 75-second carbon dioxide (CO2) breath records captured with TidalSense's N-TidalTM capnometer.

METHOD

For COPD diagnosis, machine learning algorithms were trained and evaluated on 294 COPD (including GOLD stages 1-4) and 705 non-COPD participants. A logistic regression model was also trained to distinguish GOLD 1 from GOLD 4 COPD with the output probability used as an index of severity.

RESULTS

The best diagnostic model achieved an AUROC of 0.890, sensitivity of 0.771, specificity of 0.850 and positive predictive value (PPV) of 0.834. Evaluating performance on all test capnograms that were confidently ruled in or out yielded PPV of 0.930 and NPV of 0.890. The severity determination model yielded an AUROC of 0.980, sensitivity of 0.958, specificity of 0.961 and PPV of 0.958 in distinguishing GOLD 1 from GOLD 4. Output probabilities from the severity determination model produced a correlation of 0.71 with percentage predicted FEV1.

CONCLUSION

The N-TidalTM device could be used alongside interpretable machine learning as an accurate, point-of-care diagnostic test for COPD, particularly in primary care as a rapid rule-in or rule-out test. N-TidalTM also could be effective in monitoring disease progression, providing a possible alternative to spirometry for disease monitoring.

Machine diagnosis of chronic obstructive pulmonary disease using a novel fast-response capnometer

BACKGROUND

Although currently most widely used in mechanical ventilation and cardiopulmonary resuscitation, features of the carbon dioxide (CO₂) waveform produced through capnometry have been shown to correlate with V/Q mismatch, dead space volume, type of breathing pattern, and small airway obstruction. This study applied feature engineering and machine learning techniques to capnography data collected by the N-Tidal™ device across four clinical studies to build a classifier that could distinguish CO₂ recordings (capnograms) of patients with COPD from those without COPD.

METHODS

Capnography data from four longitudinal observational studies (CBRS, GBRS, CBRS2 and ABRS) was analysed from 295 patients, generating a total of 88,186 capnograms. CO₂ sensor data was processed using TidalSense's regulated cloud platform, performing real-time geometric analysis on CO₂ waveforms to generate 82 physiologic features per capnogram. These features were used to train machine learning classifiers to discriminate COPD from 'non-COPD' (a group that included healthy participants and those with other cardiorespiratory conditions); model performance was validated on independent test sets.

RESULTS

The best machine learning model (XGBoost) performance provided a class-balanced AUROC of 0.985 ± 0.013, positive predictive value (PPV) of 0.914 ± 0.039 and sensitivity of 0.915 ± 0.066 for a diagnosis of COPD. The waveform features that are most important for driving classification are related to the alpha angle and expiratory plateau regions. These features correlated with spirometry readings, supporting their proposed properties as markers of COPD.

CONCLUSION

The N-Tidal™ device can be used to accurately diagnose COPD in near-real-time, lending support to future use in a clinical setting.

TRIAL REGISTRATION

Please see NCT03615365, NCT02814253, NCT04504838 and NCT03356288.

Functional Magnetic Resonance Imaging Connectivity Accurately Distinguishes Cases With Psychotic Disorders From Healthy Controls, Based on Cortical Features Associated With Brain Network Development

BACKGROUND

Machine learning (ML) can distinguish cases with psychotic disorder from healthy controls based on magnetic resonance imaging (MRI) data, but it is not yet clear which MRI metrics are the most informative for case-control ML, or how ML algorithms relate to the underlying biology.

METHODS

We analyzed multimodal MRI data from 2 independent case-control studies of psychotic disorders (cases, n = 65, 28; controls, n = 59, 80) and compared ML accuracy across 5 selected MRI metrics from 3 modalities. Cortical thickness, mean diffusivity, and fractional anisotropy were estimated at each of 308 cortical regions, as well as functional and structural connectivity between each pair of regions. Functional connectivity data were also used to classify nonpsychotic siblings of cases (n = 64) and to distinguish cases from controls in a third independent study (cases, n = 67; controls, n = 81).

RESULTS

In both principal studies, the most informative metric was functional MRI connectivity: The areas under the receiver operating characteristic curve were 88% and 76%, respectively. The cortical map of diagnostic connectivity features (ML weights) was replicable between studies (r = .27, p < .001); correlated with replicable case-control differences in functional MRI degree centrality and with a prior cortical map of adolescent development of functional connectivity; predicted intermediate probabilities of psychosis in siblings; and was replicated in the third case-control study.

CONCLUSIONS

ML most accurately distinguished cases from controls by a replicable pattern of functional MRI connectivity features, highlighting abnormal hubness of cortical nodes in an anatomical pattern consistent with the concept of psychosis as a disorder of network development.

Waves of Maturation and Senescence in Micro-structural MRI Markers of Human Cortical Myelination over the Lifespan

Seminal human brain histology work has demonstrated developmental waves of myelination. Here, using a micro-structural magnetic resonance imaging (MRI) marker linked to myelin, we studied fine-grained age differences to deduce waves of growth, stability, and decline of cortical myelination over the life-cycle. In 484 participants, aged 8–85 years, we fitted smooth growth curves to T1- to T2-weighted ratio in each of 360 regions from one of seven cytoarchitectonic classes. From the first derivatives of these generally inverted-U trajectories, we defined three milestones: the age at peak growth; the age at onset of a stable plateau; and the age at the onset of decline. Age at peak growth had a bimodal distribution comprising an early (pre-pubertal) wave of primary sensory and motor cortices and a later (post-pubertal) wave of association, insular and limbic cortices. Most regions reached stability in the 30-s but there was a second wave reaching stability in the 50-s. Age at onset of decline was also bimodal: in some right hemisphere regions, the curve declined from the 60-s, but in other left hemisphere regions, there was no significant decline from the stable plateau. These results are consistent with regionally heterogeneous waves of intracortical myelinogenesis and age-related demyelination.

Probabilistic thresholding of functional connectomes: Application to schizophrenia

Functional connectomes are commonly analysed as sparse graphs, constructed by thresholding cross-correlations between regional neurophysiological signals. Thresholding generally retains the strongest edges (correlations), either by retaining edges surpassing a given absolute weight, or by constraining the edge density. The latter (more widely used) method risks inclusion of false positive edges at high edge densities and exclusion of true positive edges at low edge densities. Here we apply new wavelet-based methods, which enable construction of probabilistically-thresholded graphs controlled for type I error, to a dataset of resting-state fMRI scans of 56 patients with schizophrenia and 71 healthy controls. By thresholding connectomes to fixed edge-specific P value, we found that functional connectomes of patients with schizophrenia were more dysconnected than those of healthy controls, exhibiting a lower edge density and a higher number of (dis)connected components. Furthermore, many participants' connectomes could not be built up to the fixed edge densities commonly studied in the literature (∼5-30%), while controlling for type I error. Additionally, we showed that the topological randomisation previously reported in the schizophrenia literature is likely attributable to "non-significant" edges added when thresholding connectomes to fixed density based on correlation. Finally, by explicitly comparing connectomes thresholded by increasing P value and decreasing correlation, we showed that probabilistically thresholded connectomes show decreased randomness and increased consistency across participants. Our results have implications for future analysis of functional connectivity using graph theory, especially within datasets exhibiting heterogenous distributions of edge weights (correlations), between groups or across participants.

Regional expression of the MAPT gene is associated with loss of hubs in brain networks and cognitive impairment in Parkinson disease and progressive supranuclear palsy

Abnormalities of tau protein are central to the pathogenesis of progressive supranuclear palsy, whereas haplotype variation of the tau gene MAPT influences the risk of Parkinson disease and Parkinson's disease dementia. We assessed whether regional MAPT expression might be associated with selective vulnerability of global brain networks to neurodegenerative pathology. Using task-free functional magnetic resonance imaging in progressive supranuclear palsy, Parkinson disease, and healthy subjects (n = 128), we examined functional brain networks and measured the connection strength between 471 gray matter regions. We obtained MAPT and SNCA microarray expression data in healthy subjects from the Allen brain atlas. Regional connectivity varied according to the normal expression of MAPT. The regional expression of MAPT correlated with the proportionate loss of regional connectivity in Parkinson's disease. Executive cognition was impaired in proportion to the loss of hub connectivity. These effects were not seen with SNCA, suggesting that alpha-synuclein pathology is not mediated through global network properties. The results establish a link between regional MAPT expression and selective vulnerability of functional brain networks to neurodegeneration.

A wavelet-based estimator of the degrees of freedom in denoised fMRI time series for probabilistic testing of functional connectivity and brain graphs

Connectome mapping using techniques such as functional magnetic resonance imaging (fMRI) has become a focus of systems neuroscience. There remain many statistical challenges in analysis of functional connectivity and network architecture from BOLD fMRI multivariate time series. One key statistic for any time series is its (effective) degrees of freedom, df, which will generally be less than the number of time points (or nominal degrees of freedom, N). If we know the df, then probabilistic inference on other fMRI statistics, such as the correlation between two voxel or regional time series, is feasible. However, we currently lack good estimators of df in fMRI time series, especially after the degrees of freedom of the “raw” data have been modified substantially by denoising algorithms for head movement. Here, we used a wavelet-based method both to denoise fMRI data and to estimate the (effective) df of the denoised process. We show that seed voxel correlations corrected for locally variable df could be tested for false positive connectivity with better control over Type I error and greater specificity of anatomical mapping than probabilistic connectivity maps using the nominal degrees of freedom. We also show that wavelet despiked statistics can be used to estimate all pairwise correlations between a set of regional nodes, assign a P value to each edge, and then iteratively add edges to the graph in order of increasing P. These probabilistically thresholded graphs are likely more robust to regional variation in head movement effects than comparable graphs constructed by thresholding correlations. Finally, we show that time-windowed estimates of df can be used for probabilistic connectivity testing or dynamic network analysis so that apparent changes in the functional connectome are appropriately corrected for the effects of transient noise bursts. Wavelet despiking is both an algorithm for fMRI time series denoising and an estimator of the (effective) df of denoised fMRI time series. Accurate estimation of df offers many potential advantages for probabilistically thresholding functional connectivity and network statistics tested in the context of spatially variant and non-stationary noise. Code for wavelet despiking, seed correlational testing and probabilistic graph construction is freely available to download as part of the BrainWavelet Toolbox at www.brainwavelet.org.

•

We lack good estimators of degrees of freedom (df) in denoised fMRI time series.

•

Wavelet despiking can simultaneously denoise and estimate effective df post-denoising.

•

We show Type I error controlled single-subject probabilistic inference for seed connectivity.

•

We describe new methods for building and probabilistically thresholding brain graphs.

•

We extend these methods to probabilistic inference for time-varying connectomics.

Mechanisms of gain control by voltage-gated channels in intrinsically-firing neurons

Gain modulation is a key feature of neural information processing, but underlying mechanisms remain unclear. In single neurons, gain can be measured as the slope of the current-frequency (input-output) relationship over any given range of inputs. While much work has focused on the control of basal firing rates and spike rate adaptation, gain control has been relatively unstudied. Of the limited studies on gain control, some have examined the roles of synaptic noise and passive somatic currents, but the roles of voltage-gated channels present ubiquitously in neurons have been less explored. Here, we systematically examined the relationship between gain and voltage-gated ion channels in a conductance-based, tonically-active, model neuron. Changes in expression (conductance density) of voltage-gated channels increased (Ca²⁺ channel), reduced (K⁺ channels), or produced little effect (h-type channel) on gain. We found that the gain-controlling ability of channels increased exponentially with the steepness of their activation within the dynamic voltage window (voltage range associated with firing). For depolarization-activated channels, this produced a greater channel current per action potential at higher firing rates. This allowed these channels to modulate gain by contributing to firing preferentially at states of higher excitation. A finer analysis of the current-voltage relationship during tonic firing identified narrow voltage windows at which the gain-modulating channels exerted their effects. As a proof of concept, we show that h-type channels can be tuned to modulate gain by changing the steepness of their activation within the dynamic voltage window. These results show how the impact of an ion channel on gain can be predicted from the relationship between channel kinetics and the membrane potential during firing. This is potentially relevant to understanding input-output scaling in a wide class of neurons found throughout the brain and other nervous systems.

Default mode network connectivity as a function of familial and environmental risk for psychotic disorder

Background

Research suggests that altered interregional connectivity in specific networks, such as the default mode network (DMN), is associated with cognitive and psychotic symptoms in schizophrenia. In addition, frontal and limbic connectivity alterations have been associated with trauma, drug use and urban upbringing, though these environmental exposures have never been examined in relation to DMN functional connectivity in psychotic disorder.

Methods

Resting-state functional MRI scans were obtained from 73 patients with psychotic disorder, 83 non-psychotic siblings of patients with psychotic disorder and 72 healthy controls. Posterior cingulate cortex (PCC) seed-based correlation analysis was used to estimate functional connectivity within the DMN. DMN functional connectivity was examined in relation to group (familial risk), group × environmental exposure (to cannabis, developmental trauma and urbanicity) and symptomatology.

Results

There was a significant association between group and PCC connectivity with the inferior parietal lobule (IPL), the precuneus (PCu) and the medial prefrontal cortex (MPFC). Compared to controls, patients and siblings had increased PCC connectivity with the IPL, PCu and MPFC. In the IPL and PCu, the functional connectivity of siblings was intermediate to that of controls and patients. No significant associations were found between DMN connectivity and (subclinical) psychotic/cognitive symptoms. In addition, there were no significant interactions between group and environmental exposures in the model of PCC functional connectivity.

Discussion

Increased functional connectivity in individuals with (increased risk for) psychotic disorder may reflect trait-related network alterations. The within-network “connectivity at rest” intermediate phenotype was not associated with (subclinical) psychotic or cognitive symptoms. The association between familial risk and DMN connectivity was not conditional on environmental exposure.

A wavelet method for modeling and despiking motion artifacts from resting-state fMRI time series

The impact of in-scanner head movement on functional magnetic resonance imaging (fMRI) signals has long been established as undesirable. These effects have been traditionally corrected by methods such as linear regression of head movement parameters. However, a number of recent independent studies have demonstrated that these techniques are insufficient to remove motion confounds, and that even small movements can spuriously bias estimates of functional connectivity. Here we propose a new data-driven, spatially-adaptive, wavelet-based method for identifying, modeling, and removing non-stationary events in fMRI time series, caused by head movement, without the need for data scrubbing. This method involves the addition of just one extra step, the Wavelet Despike, in standard pre-processing pipelines. With this method, we demonstrate robust removal of a range of different motion artifacts and motion-related biases including distance-dependent connectivity artifacts, at a group and single-subject level, using a range of previously published and new diagnostic measures. The Wavelet Despike is able to accommodate the substantial spatial and temporal heterogeneity of motion artifacts and can consequently remove a range of high and low frequency artifacts from fMRI time series, that may be linearly or non-linearly related to physical movements. Our methods are demonstrated by the analysis of three cohorts of resting-state fMRI data, including two high-motion datasets: a previously published dataset on children (N=22) and a new dataset on adults with stimulant drug dependence (N=40). We conclude that there is a real risk of motion-related bias in connectivity analysis of fMRI data, but that this risk is generally manageable, by effective time series denoising strategies designed to attenuate synchronized signal transients induced by abrupt head movements. The Wavelet Despiking software described in this article is freely available for download at www.brainwavelet.org.

Neuroendocrine and sympathetic responses to an orexin receptor antagonist, SB-649868, and alprazolam following insulin-induced hypoglycemia in humans

RATIONALE

The orexin-hypocretin system is important for translating peripheral metabolic signals and central neuronal inputs to a diverse range of behaviors, from feeding, motivation and arousal, to sleep and wakefulness. Orexin signaling is thus an exciting potential therapeutic target for disorders of sleep, feeding, addiction, and stress.

OBJECTIVES/METHODS

Here, we investigated the low dose pharmacology of orexin receptor antagonist, SB-649868, on neuroendocrine, sympathetic nervous system, and behavioral responses to insulin-induced hypoglycemic stress, in 24 healthy male subjects (aged 18-45 years; BMI 19.0-25.9 kg/m(2)), using a randomized, double-blind, placebo-controlled, within-subject crossover design. Alprazolam, a licensed benzodiazepine anxiolytic, was used as a positive comparator, as it has previously been validated using the insulin tolerance test (ITT) model in humans.

RESULTS

Of the primary endpoints, ITT induced defined increases in pulse rate, plasma cortisol, and adrenocorticotropic hormone in the placebo condition, but these responses were not significantly impacted by alprazolam or SB-649868 pre-treatment. Of the secondary endpoints, ITT induced a defined increase in plasma concentrations of adrenaline, noradrenaline, growth hormone (GH), and prolactin in the placebo condition. Alprazolam pre-treatment significantly reduced the GH response to ITT (p < 0.003), the peak electromyography (p < 0.0001) and galvanic skin response (GSR, p = 0.04) to acoustic startle, the resting GSR (p = 0.01), and increased appetite following ITT (p < 0.0005). SB-649868 pre-treatment produced no significant results.

CONCLUSION

We concluded that the ITT model may be informative for assessing the effects of drugs directly acting on the neuroendocrine or sympathetic nervous systems, but could not be validated for studying low dose orexin antagonist activity.

Long-term effects of attentional performance on functional brain network topology

Individuals differ in their cognitive resilience. Less resilient people demonstrate a greater tendency to vigilance decrements within sustained attention tasks. We hypothesized that a period of sustained attention is followed by prolonged changes in the organization of "resting state" brain networks and that individual differences in cognitive resilience are related to differences in post-task network reorganization. We compared the topological and spatial properties of brain networks as derived from functional MRI data (N = 20) recorded for 6 mins before and 12 mins after the performance of an attentional task. Furthermore we analysed changes in brain topology during task performance and during the switches between rest and task conditions. The cognitive resilience of each individual was quantified as the rate of increase in response latencies over the 32-minute time course of the attentional paradigm. On average, functional networks measured immediately post-task demonstrated significant and prolonged changes in network organization compared to pre-task networks with higher connectivity strength, more clustering, less efficiency, and shorter distance connections. Individual differences in cognitive resilience were significantly correlated with differences in the degree of recovery of some network parameters. Changes in network measures were still present in less resilient individuals in the second half of the post-task period (i.e. 6-12 mins after task completion), while resilient individuals already demonstrated significant reductions of functional connectivity and clustering towards pre-task levels. During task performance brain topology became more integrated with less clustering and higher global efficiency, but linearly decreased with ongoing time-on-task. We conclude that sustained attentional task performance has prolonged, "hang-over" effects on the organization of post-task resting-state brain networks; and that more cognitively resilient individuals demonstrate faster rates of network recovery following a period of attentional effort.

Human brain functional network changes associated with enhanced and impaired attentional task performance

How is the cognitive performance of the human brain related to its topological and spatial organization as a complex network embedded in anatomical space? To address this question, we used nicotine replacement and duration of attentionally demanding task performance (time-on-task), as experimental factors expected, respectively, to enhance and impair cognitive function. We measured resting-state fMRI data, performance and brain activation on a go/no-go task demanding sustained attention, and subjective fatigue in n = 18 healthy, briefly abstinent, cigarette smokers scanned repeatedly in a placebo-controlled, crossover design. We tested the main effects of drug (placebo vs Nicorette gum) and time-on-task on behavioral performance and brain functional network metrics measured in binary graphs of 477 regional nodes (efficiency, measure of integrative topology; clustering, a measure of segregated topology; and the Euclidean physical distance between connected nodes, a proxy marker of wiring cost). Nicotine enhanced attentional task performance behaviorally and increased efficiency, decreased clustering, and increased connection distance of brain networks. Greater behavioral benefits of nicotine were correlated with stronger drug effects on integrative and distributed network configuration and with greater frequency of cigarette smoking. Greater time-on-task had opposite effects: it impaired attentional accuracy, decreased efficiency, increased clustering, and decreased connection distance of networks. These results are consistent with hypothetical predictions that superior cognitive performance should be supported by more efficient, integrated (high capacity) brain network topology at greater connection distance (high cost). They also demonstrate that brain network analysis can provide novel and theoretically principled pharmacodynamic biomarkers of pro-cognitive drug effects in humans.

Tuning low-voltage-activated A-current for silent gain modulation

Modulation of stimulus-response gain and stability of spontaneous (unstimulated) firing are both important for neural computation. However, biologically plausible mechanisms that allow these distinct functional capabilities to coexist in the same neuron are poorly defined. Low-threshold, inactivating (A-type) K(+) currents (I(A)) are found in many biological neurons and are historically known for enabling low-frequency firing. By performing simulations using a conductance-based model neuron, here we show that biologically plausible shifts in I(A) conductance and inactivation kinetics produce dissociated effects on gain and intrinsic firing. This enables I(A) to regulate gain without major changes in intrinsic firing rate. Tuning I(A) properties may thus represent a previously unsuspected single-current mechanism of silent gain control in neurons.