Psilocybin modulation of time-varying functional connectivity is associated with plasma psilocin and subjective effects.

BACKGROUND: Psilocin, the neuroactive metabolite of psilocybin, is a serotonergic psychedelic that induces an acute altered state of consciousness, evokes lasting changes in mood and personality in healthy individuals, and has potential as an antidepressant treatment. Examining the acute effects of psilocin on resting-state time-varying functional connectivity implicates network-level connectivity motifs that may underlie acute and lasting behavioral and clinical effects. AIM: Evaluate the association between resting-state time-varying functional connectivity (tvFC) characteristics and plasma psilocin level (PPL) and subjective drug intensity (SDI) before and right after intake of a psychedelic dose of psilocybin in healthy humans. METHODS: Fifteen healthy individuals completed the study. Before and at multiple time points after psilocybin intake, we acquired 10-minute resting-state blood-oxygen-level-dependent functional magnetic resonance imaging scans. Leading Eigenvector Dynamics Analysis (LEiDA) and diametrical clustering were applied to estimate discrete, sequentially active brain states. We evaluated associations between the fractional occurrence of brain states during a scan session and PPL and SDI using linear mixed-effects models. We examined associations between brain state dwell time and PPL and SDI using frailty Cox proportional hazards survival analysis. RESULTS: Fractional occurrences for two brain states characterized by lateral frontoparietal and medial fronto-parietal-cingulate coherence were statistically significantly negatively associated with PPL and SDI. Dwell time for these brain states was negatively associated with SDI and, to a lesser extent, PPL. Conversely, fractional occurrence and dwell time of a fully connected brain state partly associated with motion was positively associated with PPL and SDI. CONCLUSION: Our findings suggest that the acute perceptual psychedelic effects induced by psilocybin may stem from drug-level associated decreases in the occurrence and duration of lateral and medial frontoparietal connectivity motifs. We apply and argue for a modified approach to modeling eigenvectors produced by LEiDA that more fully acknowledges their underlying structure. Together these findings contribute to a more comprehensive neurobiological framework underlying acute effects of serotonergic psychedelics.

Combining electro- and magnetoencephalography data using directional archetypal analysis.

Metastable microstates in electro- and magnetoencephalographic (EEG and MEG) measurements are usually determined using modified k-means accounting for polarity invariant states. However, hard state assignment approaches assume that the brain traverses microstates in a discrete rather than continuous fashion. We present multimodal, multisubject directional archetypal analysis as a scale and polarity invariant extension to archetypal analysis using a loss function based on the Watson distribution. With this method, EEG/MEG microstates are modeled using subject- and modality-specific archetypes that are representative, distinct topographic maps between which the brain continuously traverses. Archetypes are specified as convex combinations of unit norm input data based on a shared generator matrix, thus assuming that the timing of neural responses to stimuli is consistent across subjects and modalities. The input data is reconstructed as convex combinations of archetypes using a subject- and modality-specific continuous archetypal mixing matrix. We showcase the model on synthetic data and an openly available face perception event-related potential data set with concurrently recorded EEG and MEG. In synthetic and unimodal experiments, we compare our model to conventional Euclidean multisubject archetypal analysis. We also contrast our model to a directional clustering model with discrete state assignments to highlight the advantages of modeling state trajectories rather than hard assignments. We find that our approach successfully models scale and polarity invariant data, such as microstates, accounting for intersubject and intermodal variability. The model is readily extendable to other modalities ensuring component correspondence while elucidating spatiotemporal signal variability.

Uncovering Cortical Units of Processing From Multi-Layered Connectomes.

Modern diffusion and functional magnetic resonance imaging (dMRI/fMRI) provide non-invasive high-resolution images from which multi-layered networks of whole-brain structural and functional connectivity can be derived. Unfortunately, the lack of observed correspondence between the connectivity profiles of the two modalities challenges the understanding of the relationship between the functional and structural connectome. Rather than focusing on correspondence at the level of connections we presently investigate correspondence in terms of modular organization according to shared canonical processing units. We use a stochastic block-model (SBM) as a data-driven approach for clustering high-resolution multi-layer whole-brain connectivity networks and use prediction to quantify the extent to which a given clustering accounts for the connectome within a modality. The employed SBM assumes a single underlying parcellation exists across modalities whilst permitting each modality to possess an independent connectivity structure between parcels thereby imposing concurrent functional and structural units but different structural and functional connectivity profiles. We contrast the joint processing units to their modality specific counterparts and find that even though data-driven structural and functional parcellations exhibit substantial differences, attributed to modality specific biases, the joint model is able to achieve a consensus representation that well accounts for both the functional and structural connectome providing improved representations of functional connectivity compared to using functional data alone. This implies that a representation persists in the consensus model that is shared by the individual modalities. We find additional support for this viewpoint when the anatomical correspondence between modalities is removed from the joint modeling. The resultant drop in predictive performance is in general substantial, confirming that the anatomical correspondence of processing units is indeed present between the two modalities. Our findings illustrate how multi-modal integration admits consensus representations well-characterizing each individual modality despite their biases and points to the importance of multi-layered connectomes as providing supplementary information regarding the brain's canonical processing units.

Early Glycemic Control Assessment Based on Consensus CGM Metrics.

Continuous glucose monitoring (CGM) has revolutionized the world of diabetes and transformed the approach to diabetes care. In this context, an expert panel has reached consensus on clinical targets for CGM data interpretation based on eight CGM metrics. At least 70% of 14 consecutive CGM days (referred to as a period) are recommended to assess glycemic control based on the metrics. In clinical practice less CGM data may be available. Therefore, the primary aim of this study is to explore the ability to recover the consensus metrics utilizing less than 14 days of CGM data (intra-period). As a secondary aim, we investigate the recovery considering two consecutive periods (inter-period). The analyses are based on real-world CGM data from 484 diabetes users (4726 periods) acquired from the Cornerstones4Care® Powered by Glooko app. Using up to 14 accumulated days, the consensus metrics are calculated for each user and period, and compared to the fully 14 accumulated intra- and inter-period days. Relatively low deviations were observed for time in range (TIR) and average based metrics when using less than 14 days, however, we observed large deviations in metrics characterizing infrequent events such as time below range (TBR). Furthermore, the consensus metrics obtained in two consecutive 14 day periods have clear discrepancies (inter-period). Recovering consensus metrics using less than 14 days might still be valuable in terms of interpreting CGM data in certain clinical contexts. However, caution should be taken if treatment decisions would be made with less than 14 days of data on critical metrics such as TBR, since the metrics characterizing infrequent events deviate substantially when less data are available. Substantial deviation is also seen when comparing across two consecutive periods, which means that care should be taken not to over-generalize consensus metric based glycemic control conclusions from one period to subsequent periods.

Using connectomics for predictive assessment of brain parcellations.

The organization of the human brain remains elusive, yet is of great importance to the mechanisms of integrative brain function. At the macroscale, its structural and functional interpretation is conventionally assessed at the level of cortical units. However, the definition and validation of such cortical parcellations are problematic due to the absence of a true gold standard. We propose a framework for quantitative evaluation of brain parcellations via statistical prediction of connectomics data. Specifically, we evaluate the extent in which the network representation at the level of cortical units (defined as parcels) accounts for high-resolution brain connectivity. Herein, we assess the pertinence and comparative ranking of ten existing parcellation atlases to account for functional (FC) and structural connectivity (SC) data based on data from the Human Connectome Project (HCP), and compare them to data-driven as well as spatially-homogeneous geometric parcellations including geodesic parcellations with similar size distributions as the atlases. We find substantial discrepancy in parcellation structures that well characterize FC and SC and differences in what well represents an individual's functional connectome when compared against the FC structure that is preserved across individuals. Surprisingly, simple spatial homogenous parcellations generally provide good representations of both FC and SC, but are inferior when their within-parcellation distribution of individual parcel sizes is matched to that of a valid atlas. This suggests that the choice of fine grained and coarse representations used by existing atlases are important. However, we find that resolution is more critical than the exact border location of parcels.

Machine Learning-Based Adherence Detection of Type 2 Diabetes Patients on Once-Daily Basal Insulin Injections.

BACKGROUND: Lack of treatment adherence can lead to life-threatening health complications for people with type 2 diabetes (T2D). Recent improvements and availability in continuous glucose monitoring (CGM) technology have enabled various possibilities to monitor diabetes treatment. Detection of missed once-daily basal insulin injections can be used to provide feedback to patients, thus improving their diabetes management. In this study, we explore how machine learning (ML) based on CGM data can be used for detecting adherence to once-daily basal insulin injections. METHODS: In-silico CGM data were generated to simulate a cohort of T2D patients on once-daily insulin injection (Tresiba®). Deep learning methods within ML based on automatic feature extraction including convolutional neural networks were explored and compared with simple feature-engineered ML classification models for adherence detection. It was further investigated whether fused expert-dependent and automatically learned features could improve performance, resulting in a comparison of six different detection models. Adherence was detected throughout each day with an increasing amount of CGM data available. RESULTS: The adherence detection accuracy improved as more CGM data became available on the day of classification. The three classification models based on expert-engineered features obtained mean accuracies of 78.6%, 78.2%, and 78.3%. The classification model based purely on learned features obtained a mean accuracy of 79.7%. The two classification models fusing expert-engineered and learned features obtained mean accuracies of 79.7% and 79.8%. All the mentioned results were obtained 16 hours after time of injection. CONCLUSION: The results suggest that adherence detection based on CGM data is feasible. Even though our study based on in-silico data indicates only slightly improved performance of more complex models, the question remains whether advanced models would outperform the simple in a real-world setting. Thus, future studies on adherence monitoring using real CGM data are relevant.

The Bayesian Cut.

An important task in the analysis of graphs is separating nodes into densely connected groups with little interaction between each other. Prominent methods here include flow based graph cutting procedures as well as statistical network modeling approaches. However, adequately accounting for this, the so-called community structure, in complex networks remains a major challenge. We present a novel generic Bayesian probabilistic model for graph cutting in which we derive an analytical solution to the marginalization of nuisance parameters under constraints enforcing community structure. As a part of the solution a large scale approximation for integrals involving multiple incomplete gamma functions is derived. Our multiple cluster solution presents a generic tool for Bayesian inference on Poisson weighted graphs across different domains. Applied on three real world social networks as well as three image segmentation problems our approach shows on par or better performance to existing spectral graph cutting and community detection methods, while learning the underlying parameter space. The developed procedure provides a principled statistical framework for graph cutting and the Bayesian Cut source code provided enables easy adoption of the procedure as an alternative to existing graph cutting methods.

Short Term Blood Glucose Prediction based on Continuous Glucose Monitoring Data.

Continuous Glucose Monitoring (CGM) has enabled important opportunities for diabetes management. This study explores the use of CGM data as input for digital decision support tools. We investigate how Recurrent Neural Networks (RNNs) can be used for Short Term Blood Glucose (STBG) prediction and compare the RNNs to conventional time-series forecasting using Autoregressive Integrated Moving Average (ARIMA). A prediction horizon up to 90 min into the future is considered. In this context, we evaluate both population-based and patient-specific RNNs and contrast them to patient-specific ARIMA models and a simple baseline predicting future observations as the last observed. We find that the population-based RNN model is the best performing model across the considered prediction horizons without the need of patient-specific data. This demonstrates the potential of RNNs for STBG prediction in diabetes patients towards detecting/mitigating severe events in the STBG, in particular hypoglycemic events. However, further studies are needed in regards to the robustness and practical use of the investigated STBG prediction models.

Changes in the left temporal microstate are a sign of cognitive decline in patients with Alzheimer's disease.

INTRODUCTION: Large-scale brain networks are disrupted in the early stages of Alzheimer's disease (AD). Electroencephalography microstate analysis, a promising method for studying brain networks, parses EEG signals into topographies representing discrete, sequential network activations. Prior studies indicate that patients with AD show a pattern of global microstate disorganization. We investigated whether any specific microstate changes could be found in patients with AD and mild cognitive impairment (MCI) compared to healthy controls (HC). MATERIALS AND METHODS: Standard EEGs were obtained from 135 HC, 117 patients with MCI, and 117 patients with AD from six Nordic memory clinics. We parsed the data into four archetypal microstates. RESULTS: There was significantly increased duration, occurrence, and coverage of microstate A in patients with AD and MCI compared to HC. When looking at microstates in specific frequency bands, we found that microstate A was affected in delta (1-4 Hz), theta (4-8 Hz), and beta (13-30 Hz), while microstate D was affected only in the delta and theta bands. Microstate features were able to separate HC from AD with an accuracy of 69.8% and HC from MCI with an accuracy of 58.7%. CONCLUSIONS: Further studies are needed to evaluate whether microstates represent a valuable disease classifier. Overall, patients with AD and MCI, as compared to HC, show specific microstate alterations, which are limited to specific frequency bands. These alterations suggest disruption of large-scale cortical networks in AD and MCI, which may be limited to specific frequency bands.

Validation of structural brain connectivity networks: The impact of scanning parameters.

Evaluation of the structural connectivity (SC) of the brain based on tractography has mainly focused on the choice of diffusion model, tractography algorithm, and their respective parameter settings. Here, we systematically validate SC derived from a post mortem monkey brain, while varying key acquisition parameters such as the b-value, gradient angular resolution and image resolution. As gold standard we use the connectivity matrix obtained invasively with histological tracers by Markov et al. (2014). As performance metric, we use cross entropy as a measure that enables comparison of the relative tracer labeled neuron counts to the streamline counts from tractography. We find that high angular resolution and high signal-to-noise ratio are important to estimate SC, and that SC derived from low image resolution (1.03 mm3) are in better agreement with the tracer network, than those derived from high image resolution (0.53 mm3) or at an even lower image resolution (2.03 mm3). In contradiction, sensitivity and specificity analyses suggest that if the angular resolution is sufficient, the balanced compromise in which sensitivity and specificity are identical remains 60-64% regardless of the other scanning parameters. Interestingly, the tracer graph is assumed to be the gold standard but by thresholding, the balanced compromise increases to 70-75%. Hence, by using performance metrics based on binarized tracer graphs, one risks losing important information, changing the performance of SC graphs derived by tractography and their dependence of different scanning parameters.

Oscillatory connectivity as a diagnostic marker of dementia due to Alzheimer's disease.

OBJECTIVE: Quantitative EEG power has not been as effective in discriminating between healthy aging and Alzheimer's disease as conventional biomarkers. But EEG coherence has shown promising results in small samples. The overall aim was to evaluate if EEG connectivity markers can discriminate between Alzheimer's disease, mild cognitive impairment, and healthy aging and to explore the early underlying changes in coherence. METHODS: EEGs were included in the analysis from 135 healthy controls, 117 patients with mild cognitive impairment, and 117 patients with Alzheimer's disease from six Nordic memory clinics. Principal component analysis was performed before multinomial regression. RESULTS: We found classification accuracies of above 95% based on coherence, imaginary part of coherence, and the weighted phase-lag index. The most prominent changes in coherence were decreased alpha coherence in Alzheimer's disease, which was correlated to the scores of the 10-word test in the Consortium to Establish a Registry for Alzheimer's Disease battery. CONCLUSIONS: The diagnostic accuracies for EEG connectivity measures are higher than findings from studies investigating EEG power and conventional Alzheimer's disease biomarkers. Furthermore, decreased alpha coherence is one of the earliest changes in Alzheimer's disease and associated with memory function. SIGNIFICANCE: EEG connectivity measures may be useful supplementary diagnostic classifiers.

Individuals with 22q11.2 deletion syndrome show intact prediction but reduced adaptation in responses to repeated sounds: Evidence from Bayesian mapping.

One of the most common copy number variants, the 22q11.2 microdeletion, confers an increased risk for schizophrenia. Since schizophrenia has been associated with an aberrant neural response to repeated stimuli through both reduced adaptation and prediction, we here hypothesized that this may also be the case in nonpsychotic individuals with a 22q11.2 deletion. We recorded high-density EEG from 19 individuals with 22q11.2 deletion syndrome (12-25 years), as well as 27 healthy volunteers with comparable age and sex distribution, while they listened to a sequence of sounds arranged in a roving oddball paradigm. Using posterior probability maps and dynamic causal modelling we tested three different models accounting for repetition dependent changes in cortical responses as well as in effective connectivity; namely an adaptation model, a prediction model, and a model including both adaptation and prediction. Repetition-dependent changes were parametrically modulated by a combination of adaptation and prediction and were apparent in both cortical responses and in the underlying effective connectivity. This effect was reduced in individuals with a 22q11.2 deletion and was negatively correlated with negative symptom severity. Follow-up analysis showed that the reduced effect of the combined adaptation and prediction model seen in individuals with 22q11.2 deletion was driven by reduced adaptation rather than prediction failure. Our findings suggest that adaptation is reduced in individuals with a 22q11.2 deletion, which can be interpreted in light of the framework of predictive coding as a failure to suppress prediction errors.

Fast Assessment of Glycemic Control based on Continuous Glucose Monitoring Data.

Diabetes has become a major public health problem in the world. In this context, early assessment of glycemic control is essential in order to avoid life-threatening health complications. A panel of diabetes experts have recently proposed a list of recommendations when using Continuous Glucose Monitoring (CGM) for glycemic control assessment including a minimum of two weeks of CGM data. A recent study has further introduced a metric called Glucose Profile Indicator (GPI) for CGM based diabetes management including a subset of the recommended CGM metrics. In this pilot study, it was investigated if less than two weeks of CGM data would impact the performance of GPI compared to the proposed two weeks of CGM data. Furthermore, logistic regression (LR) was used to examine if an improvement could be achieved taking as input the CGM metrics used to quantify GPI. The population mean accuracy for accumulated day 1 to 13 varied between 72.8 ± 2.0% - 98.3 ± 0.4% with no clear sign of improvement using LR. Hence, this indicates a trade-off between the amount of available CGM data and the precision in which the GPI outcome using all 14 days can be achieved when considering features of the GPI alone. Future work is needed to investigate if this trade-off can be improved by the use of additional features of the CGM.

EEG Theta Power Is an Early Marker of Cognitive Decline in Dementia due to Alzheimer's Disease.

BACKGROUND: Quantitative EEG (qEEG) power could potentially be used as a diagnostic tool for Alzheimer's disease (AD) and may further our understanding of the pathophysiology. However, the early qEEG power changes of AD are not well understood. OBJECTIVE: To investigate the early changes in qEEG power and the possible correlation with memory function and cerebrospinal fluid biomarkers. In addition, whether qEEG power could discriminate between AD, mild cognitive impairment (MCI), and older healthy controls (HC) at the individual level. METHODS: Standard EEGs from 138 HC, 117 MCI, and 117 AD patients were included from six Nordic memory clinics. All EEGs were recorded consecutively before the diagnosis and were not used for the consensus diagnosis. Absolute and relative power was calculated for both eyes closed and open condition. RESULTS: At group level using relative power, we found significant increases globally in the theta band and decreases in high frequency power in the temporal regions for eyes closed for AD and, to a lesser extent, for MCI compared to HC. Relative theta power was significantly correlated with multiple neuropsychological measures and had the largest correlation coefficient with total tau. At the individual level, the classification rate for AD and HC was 72.9% for relative power with eyes closed. CONCLUSION: Our findings suggest that the increase in relative theta power may be the first change in patients with dementia due to AD. At the individual level, we found a moderate classification rate for AD and HC when using EEGs alone.

Rigorous optimisation of multilinear discriminant analysis with Tucker and PARAFAC structures.

BACKGROUND: We propose rigorously optimised supervised feature extraction methods for multilinear data based on Multilinear Discriminant Analysis (MDA) and demonstrate their usage on Electroencephalography (EEG) and simulated data. While existing MDA methods use heuristic optimisation procedures based on an ambiguous Tucker structure, we propose a rigorous approach via optimisation on the cross-product of Stiefel manifolds. We also introduce MDA methods with the PARAFAC structure. We compare the proposed approaches to existing MDA methods and unsupervised multilinear decompositions. RESULTS: We find that manifold optimisation substantially improves MDA objective functions relative to existing methods and on simulated data in general improve classification performance. However, we find similar classification performance when applied to the electroencephalography data. Furthermore, supervised approaches substantially outperform unsupervised mulitilinear methods whereas methods with the PARAFAC structure perform similarly to those with Tucker structures. Notably, despite applying the MDA procedures to raw Brain-Computer Interface data, their performances are on par with results employing ample pre-processing and they extract discriminatory patterns similar to the brain activity known to be elicited in the investigated EEG paradigms. CONCLUSION: The proposed usage of manifold optimisation constitutes the first rigorous and monotonous optimisation approach for MDA methods and allows for MDA with the PARAFAC structure. Our results show that MDA methods applied to raw EEG data can extract discriminatory patterns when compared to traditional unsupervised multilinear feature extraction approaches, whereas the proposed PARAFAC structured MDA models provide meaningful patterns of activity.

Altered auditory processing and effective connectivity in 22q11.2 deletion syndrome.

22q11.2 deletion syndrome (22q11.2DS) is one of the most common copy number variants and confers a markedly increased risk for schizophrenia. As such, 22q11.2DS is a homogeneous genetic liability model which enables studies to delineate functional abnormalities that may precede disease onset. Mismatch negativity (MMN), a brain marker of change detection, is reduced in people with schizophrenia compared to healthy controls. Using dynamic causal modelling (DCM), previous studies showed that top-down effective connectivity linking the frontal and temporal cortex is reduced in schizophrenia relative to healthy controls in MMN tasks. In the search for early risk-markers for schizophrenia we investigated the neural basis of change detection in a group with 22q11.2DS. We recorded high-density EEG from 19 young non-psychotic 22q11.2 deletion carriers, as well as from 27 healthy non-carriers with comparable age distribution and sex ratio, while they listened to a sequence of sounds arranged in a roving oddball paradigm. Despite finding no significant reduction in the MMN responses, whole-scalp spatiotemporal analysis of responses to the tones revealed a greater fronto-temporal N1 component in the 22q11.2 deletion carriers. DCM showed reduced intrinsic connection within right primary auditory cortex as well as in the top-down, connection from the right inferior frontal gyrus to right superior temporal gyrus for 22q11.2 deletion carriers although not surviving correction for multiple comparison. We discuss these findings in terms of reduced adaptation and a general increased sensitivity to tones in 22q11.2DS.

Predictive assessment of models for dynamic functional connectivity.

In neuroimaging, it has become evident that models of dynamic functional connectivity (dFC), which characterize how intrinsic brain organization changes over time, can provide a more detailed representation of brain function than traditional static analyses. Many dFC models in the literature represent functional brain networks as a meta-stable process with a discrete number of states; however, there is a lack of consensus on how to perform model selection and learn the number of states, as well as a lack of understanding of how different modeling assumptions influence the estimated state dynamics. To address these issues, we consider a predictive likelihood approach to model assessment, where models are evaluated based on their predictive performance on held-out test data. Examining several prominent models of dFC (in their probabilistic formulations) we demonstrate our framework on synthetic data, and apply it on two real-world examples: a face recognition EEG experiment and resting-state fMRI. Our results evidence that both EEG and fMRI are better characterized using dynamic modeling approaches than by their static counterparts, but we also demonstrate that one must be cautious when interpreting dFC because parameter settings and modeling assumptions, such as window lengths and emission models, can have a large impact on the estimated states and consequently on the interpretation of the brain dynamics.

22q11.2 Deletion Syndrome Is Associated With Impaired Auditory Steady-State Gamma Response.

Background: The 22q11.2 deletion syndrome confers a markedly increased risk for schizophrenia. 22q11.2 deletion carriers without manifest psychotic disorder offer the possibility to identify functional abnormalities that precede clinical onset. Since schizophrenia is associated with a reduced cortical gamma response to auditory stimulation at 40 Hz, we hypothesized that the 40 Hz auditory steady-state response (ASSR) may be attenuated in nonpsychotic individuals with a 22q11.2 deletion. Methods: Eighteen young nonpsychotic 22q11.2 deletion carriers and a control group of 27 noncarriers with comparable age range (12-25 years) and sex ratio underwent 128-channel EEG. We recorded the cortical ASSR to a 40 Hz train of clicks, given either at a regular inter-stimulus interval of 25 ms or at irregular intervals jittered between 11 and 37 ms. Results: Healthy noncarriers expressed a stable ASSR to regular but not in the irregular 40 Hz click stimulation. Both gamma power and inter-trial phase coherence of the ASSR were markedly reduced in the 22q11.2 deletion group. The ability to phase lock cortical gamma activity to regular auditory 40 Hz stimulation correlated with the individual expression of negative symptoms in deletion carriers (ρ = -0.487, P = .041). Conclusions: Nonpsychotic 22q11.2 deletion carriers lack efficient phase locking of evoked gamma activity to regular 40 Hz auditory stimulation. This abnormality indicates a dysfunction of fast intracortical oscillatory processing in the gamma-band. Since ASSR was attenuated in nonpsychotic deletion carriers, ASSR deficiency may constitute a premorbid risk marker of schizophrenia.

A deep learning approach to adherence detection for type 2 diabetics.

Diabetes has become one of the biggest health problems in the world. In this context, adherence to insulin treatment is essential in order to avoid life-threatening complications. In this pilot study, a novel adherence detection algorithm using Deep Learning (DL) approaches was developed for type 2 diabetes (T2D) patients, based on simulated Continuous Glucose Monitoring (CGM) signals. A large and diverse amount of CGM signals were simulated for T2D patients using a T2D adapted version of the Medtronic Virtual Patient (MVP) model for T1D. By using these signals, different classification algorithms were compared using a comprehensive grid search. We contrast a standard logistic regression baseline to Multi- Layer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs). The best classification performance with an average accuracy of 77:5% was achieved with CNN. Hence, this indicates the potential of DL, when considering adherence detection systems for T2D patients.