Using Machine Learning-Based Algorithms to Identify and Quantify Exercise Limitations in Clinical Practice: Are We There Yet?

INTRODUCTION: Well-trained staff is needed to interpret cardiopulmonary exercise tests (CPET). We aimed to examine the accuracy of machine learning-based algorithms to classify exercise limitations and their severity in clinical practice compared with expert consensus using patients presenting at a pulmonary clinic. METHODS: This study included 200 historical CPET data sets (48.5% female) of patients older than 40 yr referred for CPET because of unexplained dyspnea, preoperative examination, and evaluation of therapy progress. Data sets were independently rated by experts according to the severity of pulmonary-vascular, mechanical-ventilatory, cardiocirculatory, and muscular limitations using a visual analog scale. Decision trees and random forests analyses were calculated. RESULTS: Mean deviations between experts in the respective limitation categories ranged from 1.0 to 1.1 points (SD, 1.2) before consensus. Random forests identified parameters of particular importance for detecting specific constraints. Central parameters were nadir ventilatory efficiency for CO 2 , ventilatory efficiency slope for CO 2 (pulmonary-vascular limitations); breathing reserve, forced expiratory volume in 1 s, and forced vital capacity (mechanical-ventilatory limitations); and peak oxygen uptake, O 2 uptake/work rate slope, and % change of the latter (cardiocirculatory limitations). Thresholds differentiating between different limitation severities were reported. The accuracy of the most accurate decision tree of each category was comparable to expert ratings. Finally, a combined decision tree was created quantifying combined system limitations within one patient. CONCLUSIONS: Machine learning-based algorithms may be a viable option to facilitate the interpretation of CPET and identify exercise limitations. Our findings may further support clinical decision making and aid the development of standardized rating instruments.

Machine Learning Facial Emotion Classifiers in Psychotherapy Research: A Proof-of-Concept Study.

BACKGROUND: New advances in the field of machine learning make it possible to track facial emotional expression with high resolution, including micro-expressions. These advances have promising applications for psychotherapy research, since manual coding (e.g., the Facial Action Coding System), is time-consuming. PURPOSE: We tested whether this technology can reliably identify in-session emotional expression in a naturalistic treatment setting, and how these measures relate to the outcome of psychotherapy. METHOD: We applied a machine learning emotion classifier to video material from 389 psychotherapy sessions of 23 patients with borderline personality pathology. We validated the findings with human ratings according to the Clients Emotional Arousal Scale (CEAS) and explored associations with treatment outcomes. RESULTS: Overall, machine learning ratings showed significant agreement with human ratings. Machine learning emotion classifiers, particularly the display of positive emotions (smiling and happiness), showed medium effect size on median-split treatment outcome (d = 0.3) as well as continuous improvement (r = 0.49, p < 0.05). Patients who dropped out form psychotherapy, showed significantly more neutral expressions, and generally less social smiling, particularly at the beginning of psychotherapeutic sessions. CONCLUSIONS: Machine learning classifiers are a highly promising resource for research in psychotherapy. The results highlight differential associations of displayed positive and negative feelings with treatment outcomes. Machine learning emotion recognition may be used for the early identification of drop-out risks and clinically relevant interactions in psychotherapy.

Aberrant Hierarchical Prediction Errors Are Associated With Transition to Psychosis: A Computational Single-Trial Analysis of the Mismatch Negativity.

BACKGROUND: Mismatch negativity reductions are among the most reliable biomarkers for schizophrenia and have been associated with increased risk for conversion to psychosis in individuals who are at clinical high risk for psychosis (CHR-P). Here, we adopted a computational approach to develop a mechanistic model of mismatch negativity reductions in CHR-P individuals and patients early in the course of schizophrenia. METHODS: Electroencephalography was recorded in 38 CHR-P individuals (15 converters), 19 patients early in the course of schizophrenia (≤5 years), and 44 healthy control participants during three different auditory oddball mismatch negativity paradigms including 10% duration, frequency, or double deviants, respectively. We modeled sensory learning with the hierarchical Gaussian filter and extracted precision-weighted prediction error trajectories from the model to assess how the expression of hierarchical prediction errors modulated electroencephalography amplitudes over sensor space and time. RESULTS: Both low-level sensory and high-level volatility precision-weighted prediction errors were altered in CHR-P individuals and patients early in the course of schizophrenia compared with healthy control participants. Moreover, low-level precision-weighted prediction errors were significantly different in CHR-P individuals who later converted to psychosis compared with nonconverters. CONCLUSIONS: Our results implicate altered processing of hierarchical prediction errors as a computational mechanism in early psychosis consistent with predictive coding accounts of psychosis. This computational model seems to capture pathophysiological mechanisms that are relevant to early psychosis and the risk for future psychosis in CHR-P individuals and may serve as predictive biomarkers and mechanistic targets for the development of novel treatments.

The effect of lysergic acid diethylamide (LSD) on whole-brain functional and effective connectivity.

Psychedelics have emerged as promising candidate treatments for various psychiatric conditions, and given their clinical potential, there is a need to identify biomarkers that underlie their effects. Here, we investigate the neural mechanisms of lysergic acid diethylamide (LSD) using regression dynamic causal modelling (rDCM), a novel technique that assesses whole-brain effective connectivity (EC) during resting-state functional magnetic resonance imaging (fMRI). We modelled data from two randomised, placebo-controlled, double-blind, cross-over trials, in which 45 participants were administered 100 µg LSD and placebo in two resting-state fMRI sessions. We compared EC against whole-brain functional connectivity (FC) using classical statistics and machine learning methods. Multivariate analyses of EC parameters revealed predominantly stronger interregional connectivity and reduced self-inhibition under LSD compared to placebo, with the notable exception of weakened interregional connectivity and increased self-inhibition in occipital brain regions as well as subcortical regions. Together, these findings suggests that LSD perturbs the Excitation/Inhibition balance of the brain. Notably, whole-brain EC did not only provide additional mechanistic insight into the effects of LSD on the Excitation/Inhibition balance of the brain, but EC also correlated with global subjective effects of LSD and discriminated experimental conditions in a machine learning-based analysis with high accuracy (91.11%), highlighting the potential of using whole-brain EC to decode or predict subjective effects of LSD in the future.

Prediction of cognitive decline in Parkinson's disease (PD) patients with electroencephalography (EEG) connectivity characterized by time-between-phase-crossing (TBPC).

The aim of the study is to identify the dynamic change pattern of EEG to predict cognitive decline in patients with Parkinson's disease. Here we demonstrate that the quantification of synchrony-pattern changes across the scalp, measured using electroencephalography (EEG), offers an alternative approach of observing an individual's functional brain organization. This method, called "Time-Between-Phase-Crossing" (TBPC), is based on the same phenomenon as the phase-lag-index (PLI); it also considers intermittent changes in the signals of phase differences between pairs of EEG signals, but additionally analyzes dynamic connectivity changes. We used data from 75 non-demented Parkinson's disease patients and 72 healthy controls, who were followed over a period of 3 years. Statistics were calculated using connectome-based modeling (CPM) and receiver operating characteristic (ROC). We show that TBPC profiles, via the use of intermittent changes in signals of analytic phase differences of pairs of EEG signals, can be used to predict cognitive decline in Parkinson's disease (p < 0.05).

Learning Extremal Representations with Deep Archetypal Analysis.

Archetypes represent extreme manifestations of a population with respect to specific characteristic traits or features. In linear feature space, archetypes approximate the data convex hull allowing all data points to be expressed as convex mixtures of archetypes. As mixing of archetypes is performed directly on the input data, linear Archetypal Analysis requires additivity of the input, which is a strong assumption unlikely to hold e.g. in case of image data. To address this problem, we propose learning an appropriate latent feature space while simultaneously identifying suitable archetypes. We thus introduce a generative formulation of the linear archetype model, parameterized by neural networks. By introducing the distance-dependent archetype loss, the linear archetype model can be integrated into the latent space of a deep variational information bottleneck and an optimal representation, together with the archetypes, can be learned end-to-end. Moreover, the information bottleneck framework allows for a natural incorporation of arbitrarily complex side information during training. As a consequence, learned archetypes become easily interpretable as they derive their meaning directly from the included side information. Applicability of the proposed method is demonstrated by exploring archetypes of female facial expressions while using multi-rater based emotion scores of these expressions as side information. A second application illustrates the exploration of the chemical space of small organic molecules. By using different kinds of side information we demonstrate how identified archetypes, along with their interpretation, largely depend on the side information provided. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s11263-020-01390-3.

Dynamic Functional Connectivity of EEG: From Identifying Fingerprints to Gender Differences to a General Blueprint for the Brain's Functional Organization.

An individual's brain functional organization is unique and can reliably be observed using modalities such as functional magnetic resonance imaging (fMRI). Here we demonstrate that a quantification of the dynamics of functional connectivity (FC) as measured using electroencephalography (EEG) offers an alternative means of observing an individual's brain functional organization. Using data from both healthy individuals as well as from patients with Parkinson's disease (PD) (n = 103 healthy individuals, n = 57 PD patients), we show that "dynamic FC" (DFC) profiles can be used to identify individuals in a large group. Furthermore, we show that DFC profiles predict gender and exhibit characteristics shared both among individuals as well as between both hemispheres. Furthermore, DFC profile characteristics are frequency band specific, indicating that they reflect distinct processes in the brain. Our empirically derived method of DFC demonstrates the potential of studying the dynamics of the functional organization of the brain using EEG.

Silence in the psychotherapy of adolescents with borderline personality pathology.

Silence in psychotherapy has been associated with different, sometimes opposing meanings. This study investigated silence during adolescent identity treatment in adolescent patients with borderline personality pathology. A more active therapeutic approach with less silence is advised in adolescent identity treatment. It was hypothesized that a session with more silence might be negatively perceived by adolescent patients. A total of 382 sessions that involved 21 female patients were analyzed. Silence was automatically detected from audio recordings. Diarization (segmenting an audio according to speaker identity) was performed. The patient's perception of the sessions was measured with the Session Evaluation Questionnaire. The amount of silence in the different speaker-switching patterns was not independent of one other. This finding supports the hypothesis of mutual attunement of patient and therapist concerning the amount of silence in a given session. Sessions with less silence were rated as being both smoother and better. The potential implications for clinical practice are discussed. The investigation of turn-taking and interpersonal temporal dynamics is relevant for psychotherapy research. The topic can be addressed efficiently using automated procedures. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Cognitive decline in Parkinson's disease is associated with reduced complexity of EEG at baseline.

Parkinson's disease is a neurodegenerative disorder requiring motor signs for diagnosis, but showing more widespread pathological alterations from its beginning. Compared to age-matched healthy individuals, patients with Parkinson's disease bear a 6-fold lifetime risk of dementia. For individualized counselling and treatment, prognostic biomarkers for assessing future cognitive deterioration in early stages of Parkinson's disease are needed. In a case-control study, 42 cognitively normal patients with Parkinson's disease were compared with 24 healthy control participants matched for age, sex and education. Tsallis entropy and band power of the δ, θ, α, ß and γ-band were evaluated in baseline EEG at eyes-open and eyes-closed condition. As the θ-band showed the most pronounced differences between Parkinson's disease and healthy control groups, further analysis focussed on this band. Tsallis entropy was then compared across groups with 16 psychological test scores at baseline and follow-ups at 6 months and 3 years. In group comparison, patients with Parkinson's disease showed lower Tsallis entropy than healthy control participants. Cognitive deterioration at 3 years was correlated with Tsallis entropy in the eyes-open condition (P < 0.00079), whereas correlation at 6 months was not yet significant. Tsallis entropy measured in the eyes-closed condition did not correlate with cognitive outcome. In conclusion, the lower the EEG entropy levels at baseline in the eyes-open condition, the higher the probability of cognitive decline over 3 years. This makes Tsallis entropy a candidate prognostic biomarker for dementia in Parkinson's disease. The ability of the cortex to execute complex functions underlies cognitive health, whereas cognitive decline might clinically appear when compensatory capacity is exhausted.

On the Difference between the Information Bottleneck and the Deep Information Bottleneck.

Combining the information bottleneck model with deep learning by replacing mutual information terms with deep neural nets has proven successful in areas ranging from generative modelling to interpreting deep neural networks. In this paper, we revisit the deep variational information bottleneck and the assumptions needed for its derivation. The two assumed properties of the data, X and Y, and their latent representation T, take the form of two Markov chains T - X - Y and X - T - Y . Requiring both to hold during the optimisation process can be limiting for the set of potential joint distributions P ( X , Y , T ) . We, therefore, show how to circumvent this limitation by optimising a lower bound for the mutual information between T and Y: I ( T ; Y ) , for which only the latter Markov chain has to be satisfied. The mutual information I ( T ; Y ) can be split into two non-negative parts. The first part is the lower bound for I ( T ; Y ) , which is optimised in deep variational information bottleneck (DVIB) and cognate models in practice. The second part consists of two terms that measure how much the former requirement T - X - Y is violated. Finally, we propose interpreting the family of information bottleneck models as directed graphical models, and show that in this framework, the original and deep information bottlenecks are special cases of a fundamental IB model.

Information Bottleneck for Estimating Treatment Effects with Systematically Missing Covariates.

Estimating the effects of an intervention from high-dimensional observational data is a challenging problem due to the existence of confounding. The task is often further complicated in healthcare applications where a set of observations may be entirely missing for certain patients at test time, thereby prohibiting accurate inference. In this paper, we address this issue using an approach based on the information bottleneck to reason about the effects of interventions. To this end, we first train an information bottleneck to perform a low-dimensional compression of covariates by explicitly considering the relevance of information for treatment effects. As a second step, we subsequently use the compressed covariates to perform a transfer of relevant information to cases where data are missing during testing. In doing so, we can reliably and accurately estimate treatment effects even in the absence of a full set of covariate information at test time. Our results on two causal inference benchmarks and a real application for treating sepsis show that our method achieves state-of-the-art performance, without compromising interpretability.

Heritability of the HIV-1 reservoir size and decay under long-term suppressive ART.

The HIV-1 reservoir is the major hurdle to curing HIV-1. However, the impact of the viral genome on the HIV-1 reservoir, i.e. its heritability, remains unknown. We investigate the heritability of the HIV-1 reservoir size and its long-term decay by analyzing the distribution of those traits on viral phylogenies from both partial-pol and viral near full-length genome sequences. We use a unique nationwide cohort of 610 well-characterized HIV-1 subtype-B infected individuals on suppressive ART for a median of 5.4 years. We find that a moderate but significant fraction of the HIV-1 reservoir size 1.5 years after the initiation of ART is explained by genetic factors. At the same time, we find more tentative evidence for the heritability of the long-term HIV-1 reservoir decay. Our findings indicate that viral genetic factors contribute to the HIV-1 reservoir size and hence the infecting HIV-1 strain may affect individual patients' hurdle towards a cure.

Host Genomics of the HIV-1 Reservoir Size and Its Decay Rate During Suppressive Antiretroviral Treatment.

BACKGROUND: The primary hurdle for the eradication of HIV-1 is the establishment of a latent viral reservoir early after primary infection. Here, we investigated the potential influence of human genetic variation on the HIV-1 reservoir size and its decay rate during suppressive antiretroviral treatment. SETTING: Genome-wide association study and exome sequencing study to look for host genetic determinants of HIV-1 reservoir measurements in patients enrolled in the Swiss HIV Cohort Study, a nation-wide prospective observational study. METHODS: We measured total HIV-1 DNA in peripheral blood mononuclear cells from study participants, as a proxy for the reservoir size at 3 time points over a median of 5.4 years, and searched for associations between human genetic variation and 2 phenotypic readouts: the reservoir size at the first time point and its decay rate over the study period. We assessed the contribution of common genetic variants using genome-wide genotyping data from 797 patients with European ancestry enrolled in the Swiss HIV Cohort Study and searched for a potential impact of rare variants and exonic copy number variants using exome sequencing data generated in a subset of 194 study participants. RESULTS: Genome-wide and exome-wide analyses did not reveal any significant association with the size of the HIV-1 reservoir or its decay rate on suppressive antiretroviral treatment. CONCLUSIONS: Our results point to a limited influence of human genetics on the size of the HIV-1 reservoir and its long-term dynamics in successfully treated individuals.

Supervised Speaker Diarization Using Random Forests: A Tool for Psychotherapy Process Research.

Speaker diarization is the practice of determining who speaks when in audio recordings. Psychotherapy research often relies on labor intensive manual diarization. Unsupervised methods are available but yield higher error rates. We present a method for supervised speaker diarization based on random forests. It can be considered a compromise between commonly used labor-intensive manual coding and fully automated procedures. The method is validated using the EMRAI synthetic speech corpus and is made publicly available. It yields low diarization error rates (M: 5.61%, STD: 2.19). Supervised speaker diarization is a promising method for psychotherapy research and similar fields.

EEG Slowing and Axial Motor Impairment Are Independent Predictors of Cognitive Worsening in a Three-Year Cohort of Patients With Parkinson's Disease.

Objective: We aimed to determine whether the combination of two parameters: (a) score of axial impairment and limb rigidity (SAILR) with (b) EEG global relative median power in the frequency range theta 4-8 Hz (GRMPT) predicted cognitive outcome in patients with Parkinson's disease (PD) better than each of these measures alone. Methods: 47 non-demented patients with PD were examined and re-examined after 3 years. At both time-points, the patients underwent a comprehensive neuropsychological and neurological assessment and EEG in eyes-closed resting-state condition. The results of cognitive tests were normalized and individually summarized to obtain a "global cognitive score" (GCS). Change of GCS was used to represent cognitive changes over time. GRMPT and SAILR was used for further analysis. Linear regression models were calculated. Results: GRMPT and SAILR independently predicted cognitive change. Combination of GRMPT and SAILR improved the significance of the regression model as compared to using each of these measures alone. GRMPT and SAILR only slightly correlate between each other. Conclusion: The combination of axial signs and rigidity with quantitative EEG improves early identification of patients with PD prone to severe cognitive decline. GRMPT and SAILR seem to reflect different disease mechanisms. Significance Combination of EEG and axial motor impairment assessment may be a valuable marker in the cognitive prognosis of PD.

Phase lag index and spectral power as QEEG features for identification of patients with mild cognitive impairment in Parkinson's disease.

OBJECTIVES: To identify quantitative EEG frequency and connectivity features (Phase Lag Index) characteristic of mild cognitive impairment (MCI) in Parkinson's disease (PD) patients and to investigate if these features correlate with cognitive measures of the patients. METHODS: We recorded EEG data for a group of PD patients with MCI (n = 27) and PD patients without cognitive impairment (n = 43) using a high-resolution recording system. The EEG files were processed and 66 frequency along with 330 connectivity (phase lag index, PLI) measures were calculated. These measures were used to classify MCI vs. MCI-free patients. We also assessed correlations of these features with cognitive tests based on comprehensive scores (domains). RESULTS: PLI measures classified PD-MCI from non-MCI patients better than frequency measures. PLI in delta, theta band had highest importance for identifying patients with MCI. Amongst cognitive domains, we identified the most significant correlations between Memory and Theta PLI, Attention and Beta PLI. CONCLUSION: PLI is an effective quantitative EEG measure to identify PD patients with MCI. SIGNIFICANCE: We identified quantitative EEG measures which are important for early identification of cognitive decline in PD.

Determinants of HIV-1 reservoir size and long-term dynamics during suppressive ART.

The HIV-1 reservoir is the major hurdle to a cure. We here evaluate viral and host characteristics associated with reservoir size and long-term dynamics in 1,057 individuals on suppressive antiretroviral therapy for a median of 5.4 years. At the population level, the reservoir decreases with diminishing differences over time, but increases in 26.6% of individuals. Viral blips and low-level viremia are significantly associated with slower reservoir decay. Initiation of ART within the first year of infection, pretreatment viral load, and ethnicity affect reservoir size, but less so long-term dynamics. Viral blips and low-level viremia are thus relevant for reservoir and cure studies.

Improving counterfactual reasoning with kernelised dynamic mixing models.

Simulation-based approaches to disease progression allow us to make counterfactual predictions about the effects of an untried series of treatment choices. However, building accurate simulators of disease progression is challenging, limiting the utility of these approaches for real world treatment planning. In this work, we present a novel simulation-based reinforcement learning approach that mixes between models and kernel-based approaches to make its forward predictions. On two real world tasks, managing sepsis and treating HIV, we demonstrate that our approach both learns state-of-the-art treatment policies and can make accurate forward predictions about the effects of treatments on unseen patients.

Combining Kernel and Model Based Learning for HIV Therapy Selection.

We present a mixture-of-experts approach for HIV therapy selection. The heterogeneity in patient data makes it difficult for one particular model to succeed at providing suitable therapy predictions for all patients. An appropriate means for addressing this heterogeneity is through combining kernel and model-based techniques. These methods capture different kinds of information: kernel-based methods are able to identify clusters of similar patients, and work well when modelling the viral response for these groups. In contrast, model-based methods capture the sequential process of decision making, and are able to find simpler, yet accurate patterns in response for patients outside these groups. We take advantage of this information by proposing a mixture-of-experts model that automatically selects between the methods in order to assign the most appropriate therapy choice to an individual. Overall, we verify that therapy combinations proposed using this approach significantly outperform previous methods.