Electronic Nose for Improved Environmental Methane Monitoring.

Reducing emissions of the key greenhouse gas methane (CH4) is increasingly highlighted as being important to mitigate climate change. Effective emission reductions require cost-effective ways to measure CH4 to detect sources and verify that mitigation efforts work. We present here a novel approach to measure methane at atmospheric concentrations by means of a low-cost electronic nose strategy where the readings of a few sensors are combined, leading to errors down to 33 ppb and coefficients of determination, R2, up to 0.91 for in situ measurements. Data from methane, temperature, humidity, and atmospheric pressure sensors were used in customized machine learning models to account for environmental cross-effects and quantify methane in the ppm-ppb range both in indoor and outdoor conditions. The electronic nose strategy was confirmed to be versatile with improved accuracy when more reference data were supplied to the quantification model. Our results pave the way toward the use of networks of low-cost sensor systems for the monitoring of greenhouse gases.

Higher Apparent Gas Transfer Velocities for CO2 Compared to CH4 in Small Lakes.

Large greenhouse gas emissions occur via the release of carbon dioxide (CO2) and methane (CH4) from the surface layer of lakes. Such emissions are modeled from the air-water gas concentration gradient and the gas transfer velocity (k). The links between k and the physical properties of the gas and water have led to the development of methods to convert k between gases through Schmidt number normalization. However, recent observations have found that such normalization of apparent k estimates from field measurements can yield different results for CH4 and CO2. We estimated k for CO2 and CH4 from measurements of concentration gradients and fluxes in four contrasting lakes and found consistently higher (on an average 1.7 times) normalized apparent k values for CO2 than CH4. From these results, we infer that several gas-specific factors, including chemical and biological processes within the water surface microlayer, can influence apparent k estimates. We highlight the importance of accurately measuring relevant air-water gas concentration gradients and considering gas-specific processes when estimating k.

Diel variability of methane emissions from lakes.

Lakes are considered the second largest natural source of atmospheric methane (CH4). However, current estimates are still uncertain and do not account for diel variability of CH4 emissions. In this study, we performed high-resolution measurements of CH4 flux from several lakes, using an automated and sensor-based flux measurement approach (in total 4,580 measurements), and demonstrated a clear and consistent diel lake CH4 flux pattern during stratification and mixing periods. The maximum of CH4 flux were always noted between 10:00 and 16:00, whereas lower CH4 fluxes typically occurred during the nighttime (00:00-04:00). Regardless of the lake, CH4 emissions were on an average 2.4 higher during the day compared to the nighttime. Fluxes were higher during daytime on nearly 80% of the days. Accordingly, estimates and extrapolations based on daytime measurements only most likely result in overestimated fluxes, and consideration of diel variability is critical to properly assess the total lake CH4 flux, representing a key component of the global CH4 budget. Hence, based on a combination of our data and additional literature information considering diel variability across latitudes, we discuss ways to derive a diel variability correction factor for previous measurements made during daytime only.

Automated flux chamber for investigating gas flux at water-air interfaces.

Aquatic ecosystems are major sources of greenhouse gases (GHG). Representative measurements of GHG fluxes from aquatic ecosystems to the atmosphere are vital for quantitative understanding of relationships between biogeochemistry and climate. Fluxes occur at high temporal variability at diel or longer scales, which are not captured by traditional short-term deployments (often in the order of 30 min) of floating flux chambers. High temporal frequency measurements are necessary but also extremely labor intensive if manual flux chamber based methods are used. Therefore, we designed an inexpensive and easily mobile automated flux chamber (AFC) for extended deployments. The AFC was designed to measure in situ accumulation of gas in the chamber and also to collect gas samples in an array of sample bottles for subsequent analysis in the laboratory, providing two independent ways of CH(4) concentration measurements. We here present the AFC design and function together with data from initial laboratory tests and from a field deployment.

Minor impacts of rain on methane flux from hemiboreal, boreal, and subarctic lakes.

Methane (CH4) emissions (FCH4) from northern freshwater lakes are not only significant but also highly variable in time and one driver variable suggested to be important is precipitation. Rain can have various, potentially large effects on FCH4 across multiple time frames, and verifying the impact of rain on lake FCH4 is key to understand both contemporary flux regulation, and to predict future FCH4 related to possible changes in frequency and intensity of rainfall from climate change. The main objective of this study was to assess the short-term impact of typically occurring rain events with different intensity on FCH4 from various lake types located in hemiboreal, boreal, and subarctic Sweden. In spite of high time resolution automated flux measurements across different depth zones and covering numerous commonly types of rain events in northern areas, in general, no strong impact on FCH4 during and within 24 h after the rainfall could be observed. Only in deeper lake areas and during longer rain events FCH4 was weakly related to rain (R2 = 0.29, p < 0.05), where a minor FCH4 decrease during the rain was identified, suggesting that direct rainwater input, during greater rainfall, may decrease FCH4 by dilution of surface water CH4. Overall, this study indicates that typical rain events in the studied regions have minor direct short-term effects on FCH4 from northern lakes and do not enhance FCH4 from shallow and deeper parts of lakes during and up to 24-h after the rainfall. Instead, other factors such as wind speed, water temperature and pressure changes were more strongly correlated with lake FCH4.

Ensemble learning using traditional machine learning and deep neural network for diagnosis of Alzheimer's disease.

In recent years, Alzheimer's disease (AD) diagnosis using neuroimaging and deep learning has drawn great research attention. However, due to the scarcity of training neuroimaging data, many deep learning models have suffered from severe overfitting. In this study, we propose an ensemble learning framework that combines deep learning and machine learning. The deep learning model was based on a 3D-ResNet to exploit 3D structural features of neuroimaging data. Meanwhile, Extreme Gradient Boosting (XGBoost) machine learning was applied on a voxel-wise basis to draw the most significant voxel groups out of the image. The 3D-ResNet and XGBoost predictions were combined with patient demographics and cognitive test scores (Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR)) to give a final diagnosis prediction. Our proposed method was trained and validated on brain MRI brain images of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. During the training phase, multiple data augmentation methods were employed to tackle overfitting. Our test set contained only baseline scans, i.e., the first visit scans since we aimed to investigate the ability of our approach in detecting AD during the first visit of AD patients. Our 5-fold cross-validation implementation achieved an average AUC of 100% during training and 96% during testing. Using the same computer, our method was much faster in scoring a prediction, approximately 10 min, than feature extraction-based machine learning methods, which often take many hours to score a prediction. To make the prediction explainable, we visualized the brain MRI image regions that primarily affected the 3D-ResNet model's prediction via heatmap. Lastly, we observed that proper generation of test sets was critical to avoiding the data leakage issue and ensuring the validity of results.

Formation of H2 and CH4 by weathering of olivine at temperatures between 30 and 70°C.

Hydrocarbons such as CH4 are known to be formed through the Fischer-Tropsch or Sabatier type reactions in hydrothermal systems usually at temperatures above 100°C. Weathering of olivine is sometimes suggested to account for abiotic formation of CH4 through its redox lowering and water splitting properties. Knowledge about the CH4 and H2 formation processes at low temperatures is important for the research about the origin and cause of early Earth and Martian CH4 and for CO2 sequestration. We have conducted a series of low temperature, long-term weathering experiments in which we have tested the CH4 and H2 formation potential of forsteritic olivine.The results show low temperature CH4 production that is probably influenced by chromite and magnetite as catalysts. Extensive analyses of a potential CH4 source trapped in the crystal structure of the olivine showed no signs of incorporated CH4. Also, the available sources of organic carbon were not enough to support the total amount of CH4 detected in our experiments. There was also a linear relationship between silica release into solution and the net CH4 accumulation into the incubation bottle headspaces suggesting that CH4 formation under these conditions could be a qualitative indicator of olivine dissolution.It is likely that minerals such as magnetite, chromite and other metal-rich minerals found on the olivine surface catalyze the formation of CH4, because of the low temperature of the system. This may expand the range of environments plausible for abiotic CH4 formation both on Earth and on other terrestrial bodies.

EEG microstate features for schizophrenia classification.

Electroencephalography (EEG) microstate analysis is a method wherein spontaneous EEG activity is segmented at sub-second levels to analyze quasi-stable states. In particular, four archetype microstates and their features are known to reflect changes in brain state in neuropsychiatric diseases. However, previous studies have only reported differences in each microstate feature and have not determined whether microstate features are suitable for schizophrenia classification. Therefore, it is necessary to validate microstate features for schizophrenia classification. Nineteen microstate features, including duration, occurrence, and coverage as well as thirty-one conventional EEG features, including statistical, frequency, and temporal characteristics were obtained from resting-state EEG recordings of 14 patients diagnosed with schizophrenia and from 14 healthy (control) subjects. Machine-learning based multivariate analysis was used to evaluate classification performance. EEG recordings of patients and controls showed different microstate features. More importantly, when differentiating among patients and controls, EEG microstate features outperformed conventional EEG ones. The performance of the microstate features exceeded that of conventional EEG, even after optimization using recursive feature elimination. EEG microstate features applied with conventional EEG features also showed better classification performance than conventional EEG features alone. The current study is the first to validate the use of microstate features to discriminate schizophrenia, suggesting that EEG microstate features are useful for schizophrenia classification.

Quantification of stroke lesion volume using epidural EEG in a cerebral ischaemic rat model.

Precise monitoring of the brain after a stroke is essential for clinical decision making. Due to the non-invasive nature and high temporal resolution of electroencephalography (EEG), it is widely used to evaluate real-time cortical activity. In this study, we investigated the stroke-related EEG biomarkers and developed a predictive model for quantifying the structural brain damage in a focal cerebral ischaemic rat model. We enrolled 31 male Sprague-Dawley rats and randomly assigned them to mild stroke, moderate stroke, severe stroke, and control groups. We induced photothrombotic stroke targeting the right auditory cortex. We then acquired EEG signal responses to sound stimuli (frequency linearly increasing from 8 to 12 kHz with 750 ms duration). Power spectral analysis revealed a significant correlation of the relative powers of alpha, theta, delta, delta/alpha ratio, and (delta + theta)/(alpha + beta) ratio with the stroke lesion volume. The auditory evoked potential analysis revealed a significant association of amplitude and latency with stroke lesion volume. Finally, we developed a multiple regression model combining EEG predictors for quantifying the ischaemic lesion (R2 = 0.938, p value < 0.001). These findings demonstrate the potential application of EEG as a valid modality for monitoring the brain after a stroke.

EEG microstate features according to performance on a mental arithmetic task.

In this study, we hypothesized that task performance could be evaluated applying EEG microstate to mental arithmetic task. This pilot study also aimed at evaluating the efficacy of microstates as novel features to discriminate task performance. Thirty-six subjects were divided into good and poor performers, depending on how well they performed the task. Microstate features were derived from EEG recordings during resting and task states. In the good performers, there was a decrease in type C and an increase in type D features during the task compared to the resting state. Mean duration and occurrence decreased and increased, respectively. In the poor performers, occurrence of type D feature, mean duration and occurrence showed greater changes. We investigated whether microstate features were suitable for task performance classification and eleven features including four archetypes were selected by recursive feature elimination (RFE). The model that implemented them showed the highest classification performance for differentiating between groups. Our pilot findings showed that the highest mean Area Under Curve (AUC) was 0.831. This study is the first to apply EEG microstate features to specific cognitive tasks in healthy subjects, suggesting that EEG microstate features can reflect task achievement.

3D-Deep Learning Based Automatic Diagnosis of Alzheimer's Disease with Joint MMSE Prediction Using Resting-State fMRI.

We performed this research to 1) evaluate a novel deep learning method for the diagnosis of Alzheimer's disease (AD) and 2) jointly predict the Mini Mental State Examination (MMSE) scores of South Korean patients with AD. Using resting-state functional Magnetic Resonance Imaging (rs-fMRI) scans of 331 participants, we obtained functional 3-dimensional (3-D) independent component spatial maps for use as features in classification and regression tasks. A 3-D convolutional neural network (CNN) architecture was developed for the classification task. MMSE scores were predicted using: linear least square regression (LLSR), support vector regression, bagging-based ensemble regression, and tree regression with group independent component analysis (gICA) features. To improve MMSE regression performance, we applied feature optimization methods including least absolute shrinkage and selection operator and support vector machine-based recursive feature elimination (SVM-RFE). The mean balanced test accuracy was 85.27% for the classification of AD versus healthy controls. The medial visual, default mode, dorsal attention, executive, and auditory related networks were mainly associated with AD. The maximum clinical MMSE score prediction accuracy with the LLSR method applied on gICA combined with SVM-RFE features had the lowest root mean square error (3.27 ± 0.58) and the highest R2 value (0.63 ± 0.02). Classification of AD and healthy controls can be successfully achieved using only rs-fMRI and MMSE scores can be accurately predicted using functional independent component features. In the absence of trained clinicians, AD disease status and clinical MMSE scores can be jointly predicted using 3-D deep learning and regression learning approaches with rs-fMRI data.

Microstate functional connectivity in EEG cognitive tasks revealed by a multivariate Gaussian hidden Markov model with phase locking value.

OBJECTIVE: Tracking the spatiotemporal fast (~100 ms) transient networks remains challenging due to a limited understanding of neural activity dynamics as well as a lack of relevant sophisticated methodologies. In this study, we introduce a novel approach to identify simultaneously distinct EEG microstates and their corresponding microstate functional connectivity (µFC) networks in which each µFC network is associated with a distinguished connectivity pattern of recurrent neuronal activity. APPROACH: The introduced approach is based on a multivariate Gaussian hidden Markov model (MGHMM) to decompose the sensor-space stochastic multi-subject event-related potential (ERP) into quasi-stable EEG microstates. Raw trial segments whose time windows belong to a corresponding segmented EEG microstate are then concatenated for measuring their µFC using the time-averaged phase-locking value. Illustration of this method is evaluated with synthetic data for which ground-truth microstate dynamics are known. Furthermore, we apply the method to identify EEG microstates and corresponding µFC networks in publicly available EEG data measured from visual cognitive tasks. Finally, we compare the MGHMM method with conventional dynamic FC (dFC) approaches using clustering-based K-means and time sliding windows, which conversely segregate the macrostate FC matrices across times into 'FC-states'. MAIN RESULTS: By using the MGHMM approach, we reveal: (1) EEG microstates, (2) µFC networks, (3) the associations of EEG microstate networks and their corresponding µFC networks dynamically modulated in publicly available EEG cognitive tasks, and (4) compared dFC performances between our proposed µFC approaches and 'FC-states' segmented by clustering-based K-means and time sliding windows. SIGNIFICANCE: Evidence of significant improvements of microstate correlations (p -value < 0.05) and improved tendency of FC distinction (p -value = 0.064) over reported methods with simulated and realistic data will make this approach a preferred methodology to study dynamic brain networks and guarantee its use for further clinical applications.

Mild Cognitive Impairment Diagnosis Using Extreme Learning Machine Combined With Multivoxel Pattern Analysis on Multi-Biomarker Resting-State FMRI.

This paper proposed a classification framework that integrates hybrid multivoxel pattern analyses (MVPA) and extreme learning machine (ELM) for automated Mild Cognitive Impairment (MCI) diagnosis applied on concatenations of multi-biomarker resting-state fMRI. Given three-dimensional (3D) regional coherences and functional connectivity patterns measured during resting state, we performed 3D univariate t-tests to obtain initial univariate features which show the significant changes. To enhance discriminative patterns, we employed multivariate feature reductions using recursive feature elimination in combination with univariate t-test. The maximal amount of information changes were achieved by concatenations of multiple functional metrics. The classifications were performed by an ELM, and its efficiency was compared to SVMs. This study reported mean accuracies using 10-fold cross-validation, followed by permutation tests to assess the statistical significance of discriminative results. In diagnosis of MCI, the proposed method achieved a maximal accuracy of 97.86% (p<; 0.001) in ADNI2 cohort and thus has potentials to assist the clinicians in MCI diagnosis.