A systematic evaluation of Euclidean alignment with deep learning for EEG decoding.

Electroencephalography (EEG) signals are frequently used for various Brain-Computer Interface (BCI) tasks. While Deep Learning (DL) techniques have shown promising results, they are hindered by the substantial data requirements. By leveraging data from multiple subjects, transfer learning enables more effective training of DL models. A technique that is gaining popularity is Euclidean Alignment (EA) due to its ease of use, low computational complexity, and compatibility with Deep Learning models. However, few studies evaluate its impact on the training performance of shared and individual DL models. In this work, we systematically evaluate the effect of EA combined with DL for decoding BCI signals. We used EA to train shared models with data from multiple subjects and evaluated its transferability to new subjects. Our experimental results show that it improves decoding in the target subject by 4.33% and decreases convergence time by more than 70%. We also trained individual models for each subject to use as a majority-voting ensemble classifier. In this scenario, using EA improved the 3-model ensemble accuracy by 3.7%. However, when compared to the shared model with EA, the ensemble accuracy was 3.62% lower.

Preeclampsia Prediction Using Machine Learning and Polygenic Risk Scores From Clinical and Genetic Risk Factors in Early and Late Pregnancies.

BACKGROUND: Preeclampsia, a pregnancy-specific condition associated with new-onset hypertension after 20-weeks gestation, is a leading cause of maternal and neonatal morbidity and mortality. Predictive tools to understand which individuals are most at risk are needed. METHODS: We identified a cohort of N=1125 pregnant individuals who delivered between May 2015 and May 2022 at Mass General Brigham Hospitals with available electronic health record data and linked genetic data. Using clinical electronic health record data and systolic blood pressure polygenic risk scores derived from a large genome-wide association study, we developed machine learning (XGBoost) and logistic regression models to predict preeclampsia risk. RESULTS: Pregnant individuals with a systolic blood pressure polygenic risk score in the top quartile had higher blood pressures throughout pregnancy compared with patients within the lowest quartile systolic blood pressure polygenic risk score. In the first trimester, the most predictive model was XGBoost, with an area under the curve of 0.74. In late pregnancy, with data obtained up to the delivery admission, the best-performing model was XGBoost using clinical variables, which achieved an area under the curve of 0.91. Adding the systolic blood pressure polygenic risk score to the models did not improve the performance significantly based on De Long test comparing the area under the curve of models with and without the polygenic score. CONCLUSIONS: Integrating clinical factors into predictive models can inform personalized preeclampsia risk and achieve higher predictive power than the current practice. In the future, personalized tools can be implemented to identify high-risk patients for preventative therapies and timely intervention to improve adverse maternal and neonatal outcomes.

An Interpretable Longitudinal Preeclampsia Risk Prediction Using Machine Learning.

Background: Preeclampsia is a pregnancy-specific disease characterized by new onset hypertension after 20 weeks of gestation that affects 2-8% of all pregnancies and contributes to up to 26% of maternal deaths. Despite extensive clinical research, current predictive tools fail to identify up to 66% of patients who will develop preeclampsia. We sought to develop a tool to longitudinally predict preeclampsia risk. Methods: In this retrospective model development and validation study, we examined a large cohort of patients who delivered at six community and two tertiary care hospitals in the New England region between 02/2015 and 06/2023. We used sociodemographic, clinical diagnoses, family history, laboratory, and vital signs data. We developed eight datasets at 14, 20, 24, 28, 32, 36, 39 weeks gestation and at the hospital admission for delivery. We created linear regression, random forest, xgboost, and deep neural networks to develop multiple models and compared their performance. We used Shapley values to investigate the global and local explainability of the models and the relationships between the predictive variables. Findings: Our study population (N=120,752) had an incidence of preeclampsia of 5.7% (N=6,920). The performance of the models as measured using the area under the curve, AUC, was in the range 0.73-0.91, which was externally validated. The relationships between some of the variables were complex and non-linear; in addition, the relative significance of the predictors varied over the pregnancy. Compared to the current standard of care for preeclampsia risk stratification in the first trimester, our model would allow 48.6% more at-risk patients to be identified. Interpretation: Our novel preeclampsia prediction tool would allow clinicians to identify patients at risk early and provide personalized predictions, as well as longitudinal predictions throughout pregnancy. Funding: National Institutes of Health, Anesthesia Patient Safety Foundation. RESEARCH IN CONTEXT: Evidence before this study: Current tools for the prediction of preeclampsia are lacking as they fail to identify up to 66% of the patients who develop preeclampsia. We searched PubMed, MEDLINE, and the Web of Science from database inception to May 1, 2023, using the keywords "deep learning", "machine learning", "preeclampsia", "artificial intelligence", "pregnancy complications", and "predictive models". We identified 13 studies that employed machine learning to develop prediction models for preeclampsia risk based on clinical variables. Among these studies, six included biomarkers such as serum placental growth factor, pregnancy-associated plasma protein A, and uterine artery pulsatility index, which are not routinely available in our clinical practice; two studies were in diverse cohorts of more than 100 000 patients, and two studies developed longitudinal predictions using medical records data. However, most studies have limited depth, concerns about data leakage, overfitting, or lack of generalizability.Added value of this study: We developed a comprehensive longitudinal predictive tool based on routine clinical data that can be used throughout pregnancy to predict the risk of preeclampsia. We tested multiple types of predictive models, including machine learning and deep learning models, and demonstrated high predictive power. We investigated the changes over different time points of individual and group variables and found previously known and novel relationships between variables such as red blood cell count and preeclampsia risk.Implications of all the available evidence: Longitudinal prediction of preeclampsia using machine learning can be achieved with high performance. Implementation of an accurate predictive tool within the electronic health records can aid clinical care and identify patients at heightened risk who would benefit from aspirin prophylaxis, increased surveillance, early diagnosis, and escalation in care. These results highlight the potential of using artificial intelligence in clinical decision support, with the ultimate goal of reducing iatrogenic preterm birth and improving perinatal care.

A Methodology for a Scalable, Collaborative, and Resource-Efficient Platform, MERLIN, to Facilitate Healthcare AI Research.

Healthcare artificial intelligence (AI) holds the potential to increase patient safety, augment efficiency and improve patient outcomes, yet research is often limited by data access, cohort curation, and tools for analysis. Collection and translation of electronic health record data, live data, and real-time high-resolution device data can be challenging and time-consuming. The development of clinically relevant AI tools requires overcoming challenges in data acquisition, scarce hospital resources, and requirements for data governance. These bottlenecks may result in resource-heavy needs and long delays in research and development of AI systems. We present a system and methodology to accelerate data acquisition, dataset development and analysis, and AI model development. We created an interactive platform that relies on a scalable microservice architecture. This system can ingest 15,000 patient records per hour, where each record represents thousands of multimodal measurements, text notes, and high-resolution data. Collectively, these records can approach a terabyte of data. The platform can further perform cohort generation and preliminary dataset analysis in 2-5 minutes. As a result, multiple users can collaborate simultaneously to iterate on datasets and models in real time. We anticipate that this approach will accelerate clinical AI model development, and, in the long run, meaningfully improve healthcare delivery.

Prediction of Preeclampsia from Clinical and Genetic Risk Factors in Early and Late Pregnancy Using Machine Learning and Polygenic Risk Scores.

Background: Preeclampsia, a pregnancy-specific condition associated with new-onset hypertension after 20 weeks gestation, is a leading cause of maternal and neonatal morbidity and mortality. Predictive tools to understand which individuals are most at risk are needed. Methods: We identified a cohort of N=1,125 pregnant individuals who delivered between 05/2015-05/2022 at Mass General Brigham hospitals with available electronic health record (EHR) data and linked genetic data. Using clinical EHR data and systolic blood pressure polygenic risk scores (SBP PRS) derived from a large genome-wide association study, we developed machine learning (xgboost) and linear regression models to predict preeclampsia risk. Results: Pregnant individuals with an SBP PRS in the top quartile had higher blood pressures throughout pregnancy compared to patients within the lowest quartile SBP PRS. In the first trimester, the most predictive model was xgboost, with an area under the curve (AUC) of 0.73. Adding the SBP PRS to the models improved the performance only of the linear regression model from AUC 0.70 to 0.71; the predictive power of other models remained unchanged. In late pregnancy, with data obtained up to the delivery admission, the best performing model was xgboost using clinical variables, which achieved an AUC of 0.91. Conclusions: Integrating clinical and genetic factors into predictive models can inform personalized preeclampsia risk and achieve higher predictive power than the current practice. In the future, personalized tools can be implemented in clinical practice to identify high-risk patients for preventative therapies and timely intervention to improve adverse maternal and neonatal outcomes.

An Orchestration Platform that Puts Radiologists in the Driver's Seat of AI Innovation: a Methodological Approach.

Current AI-driven research in radiology requires resources and expertise that are often inaccessible to small and resource-limited labs. The clinicians who are able to participate in AI research are frequently well-funded, well-staffed, and either have significant experience with AI and computing, or have access to colleagues or facilities that do. Current imaging data is clinician-oriented and is not easily amenable to machine learning initiatives, resulting in inefficient, time consuming, and costly efforts that rely upon a crew of data engineers and machine learning scientists, and all too often preclude radiologists from driving AI research and innovation. We present the system and methodology we have developed to address infrastructure and platform needs, while reducing the staffing and resource barriers to entry. We emphasize a data-first and modular approach that streamlines the AI development and deployment process while providing efficient and familiar interfaces for radiologists, such that they can be the drivers of new AI innovations.

Time experience during social distancing: A longitudinal study during the first months of COVID-19 pandemic in Brazil.

Social distancing in response to the COVID-19 pandemic brought several modifications in our daily lives. With these changes, some people have reported alterations in their feelings of how fast time was passing. In this study, we assessed whether and how social distancing and the evolution of the COVID-19 pandemic influenced participants' time awareness and production of time intervals. Participants (n = 3855) filled in the first questionnaire approximately 60 days after the start of social distancing in Brazil and weekly questionnaires for 15 weeks during social distancing. Our results indicate that time was perceived as expanded at the beginning, but this feeling decreased across the weeks. Time awareness was strongly associated with psychological factors such as loneliness, stress, and positive emotions, but not with time production. This relation was shown between participants and within their longitudinal reports. Together, our findings show how emotions are a crucial aspect of how time is felt.

Monitored anesthesia care during mechanical thrombectomy for stroke: need for data-driven and individualized decisions.

BACKGROUND: The optimal anesthesia management for patients with stroke undergoing mechanical thrombectomy (MT) during the COVID-19 pandemic has become a matter of controversy. Some recent guidelines have favored general anesthesia (GA) in patients perceived as high risk for intraprocedural conversion from sedation to GA, including those with dominant hemispheric occlusions/aphasia or baseline National Institutes of Health Stroke Scale (NIHSS) score >15. We aim to identify the rate and predictors of conversion to GA during MT in a high-volume center where monitored anesthesia care (MAC) is the default modality. METHODS: A retrospective review of a prospectively maintained MT database from January 2013 to July 2020 was undertaken. Analyses were conducted to identify the predictors of intraprocedural conversion to GA. In addition, we analyzed the GA conversion rates in subgroups of interest. RESULTS: Among 1919 MT patients, 1681 (87.6%) started treatment under MAC (median age 65 years (IQR 55-76); baseline NIHSS 16 (IQR 11-21); 48.4% women). Of the 1677 eligible patients, 26 (1.6%) converted to GA including 1.4% (22/1615) with anterior and 6.5% (4/62) with posterior circulation strokes. The only predictor of GA conversion was posterior circulation stroke (OR 4.99, 95% CI 1.67 to 14.96, P=0.004). The conversion rates were numerically higher in right than in left hemispheric occlusions (1.6% vs 1.2%; OR 1.37, 95% CI 0.59 to 3.19, P=0.47) and in milder than in more severe strokes (NIHSS ≤15 vs >15: 2% vs 1.2%; OR 0.62, 95% CI 0.28 to 1.36, P=0.23). CONCLUSIONS: Our study showed that the overall rate of conversion from MAC to GA during MT was low (1.6%) and, while higher in posterior circulation strokes, it was not predicted by either hemispheric dominance or stroke severity. Caution should be given before changing clinical practice during moments of crisis.

Directed functional and structural connectivity in a large-scale model for the mouse cortex.

Inferring the structural connectivity from electrophysiological measurements is a fundamental challenge in systems neuroscience. Directed functional connectivity measures, such as the generalized partial directed coherence (GPDC), provide estimates of the causal influence between areas. However, the relation between causality estimates and structural connectivity is still not clear. We analyzed this problem by evaluating the effectiveness of GPDC to estimate the connectivity of a ground-truth, data-constrained computational model of a large-scale network model of the mouse cortex. The model contains 19 cortical areas composed of spiking neurons, with areas connected by long-range projections with weights obtained from a tract-tracing cortical connectome. We show that GPDC values provide a reasonable estimate of structural connectivity, with an average Pearson correlation over simulations of 0.74. Moreover, even in a typical electrophysiological recording scenario containing five areas, the mean correlation was above 0.6. These results suggest that it may be possible to empirically estimate structural connectivity from functional connectivity even when detailed whole-brain recordings are not achievable.

Dosimetric benefits of gantry-static couch-motion (GsCM) technique for breast boost radiation therapy: Reduced dose to organs-at-risk and improved dosimetric indices.

To evaluate the clinical feasibility and dosimetric benefits of a novel gantry-static couch-motion (GsCM) technique for external beam photon boost treatment of lumpectomy cavity in patients with early-stage breast cancer in comparison to three-dimensional conformal radiotherapy (3D-CRT), wedge pair in supine position (WPS), and wedge pair in decubitus position (WPD) techniques. A retrospective review was conducted on breast patients (right breast, n = 10 and left breast, n = 10) who received 10 Gy boost after 50 Gy to whole breast. The treatment plans were generated using an isocentric-based GsCM technique (a VMAT type planning approach) integrating couch rotational motion at static gantry positions. Static fields for each tangential side were merged using a Matlab® script and delivered automatically within the Varian TruebeamTM STx in Developer Mode application as a VMAT arc (wide-angular medial and short-angular lateral arcs). The dosimetric accuracy of the plan delivery was evaluated by ion chamber array measurements in phantom. For both right and left breast boost GsCM, 3D-CRT, WPS, and WPD all provided an adequate coverage to PTV. GsCM significantly reduced the ipsilateral lung V30% for right side (mean, 80%) and left side (mean, 70%). Heart V5% reduced by 90% (mean) for right and 80% (mean) for left side. Ipsilateral breast V50% and mean dose were comparable for all techniques but for GsCM, V100% reduced by 50% (mean) for right and left side. The automated delivery of both arcs was under 2 min as compared to delivering individual fields (30 ± 5 min). The gamma analysis using 2 mm distance to agreement (DTA) and 2% dose difference (DD) was 98 ± 1.5% for all 20 plans. The GsCM technique facilitates coronal plane dose delivery appropriate for deep-seated breast boost cavities, with sufficient dose conformity of target volume paired with sparing of the OARs.

A model for the peak-interval task based on neural oscillation-delimited states.

Specific mechanisms underlying how the brain keeps track of time are largely unknown. Several existing computational models of timing reproduce behavioral results obtained with experimental psychophysical tasks, but only a few tackle the underlying biological mechanisms, such as the synchronized neural activity that occurs throughout brain areas. In this paper, we introduce a model for the peak-interval task based on neuronal network properties. We consider that Local Field Potential (LFP) oscillation cycles specify a sequence of states, represented as neuronal ensembles. Repeated presentation of time intervals during training reinforces the connections of specific ensembles to downstream networks - sets of neurons connected to the sequence of states. Later, during the peak-interval procedure, these downstream networks are reactivated by previously experienced neuronal ensembles, triggering behavioral responses at the learned time intervals. The model reproduces experimental response patterns from individual rats in the peak-interval procedure, satisfying relevant properties such as the Weber law. Finally, we provide a biological interpretation of the parameters of the model.

Evaluation of connectivity estimates using spiking neuronal network models.

The flow of information between different regions of the cortex is fundamental for brain function. Researchers use causality detection techniques, such as Granger causality, to infer connectivity among brain areas from time series. Generalized partial directed coherence (GPDC) is a frequency domain linear method based on vector autoregressive model, which has been applied in electroencephalography, local field potential, and blood oxygenation level-dependent signals. Despite its widespread usage, previous attempts to validate GPDC use oversimplified simulated data, which do not reflect the nonlinearities and network couplings present in biological signals. In this work, we evaluated the GPDC performance when applied to simulated LFP signals, i.e., generated from networks of spiking neuronal models. We created three models, each containing five interacting networks, and evaluated whether the GPDC method could accurately detect network couplings. When using a stronger coupling, we showed that GPDC correctly detects all existing connections from simulated LFP signals in the three models, without false positives. Varying the coupling strength between networks, by changing the number of connections or synaptic strengths, and adding noise in the times series, altered the receiver operating characteristic (ROC) curve, ranging from perfect to chance level retrieval. We also showed that GPDC values correlated with coupling strength, indicating that GPDC values can provide useful information regarding coupling strength. These results reinforce that GPDC can be used to detect causality relationships over neural signals.

A common representation of time across visual and auditory modalities.

Humans' and non-human animals' ability to process time on the scale of milliseconds and seconds is essential for adaptive behaviour. A central question of how brains keep track of time is how specific temporal information across different sensory modalities is. In the present study, we show that encoding of temporal intervals in auditory and visual modalities are qualitatively similar. Human participants were instructed to reproduce intervals in the range from 750 ms to 1500 ms marked by auditory or visual stimuli. Our behavioural results suggest that, although participants were more accurate in reproducing intervals marked by auditory stimuli, there was a strong correlation in performance between modalities. Using multivariate pattern analysis in scalp EEG, we show that activity during late periods of the intervals was similar within and between modalities. Critically, we show that a multivariate pattern classifier was able to accurately predict the elapsed interval, even when trained on an interval marked by a stimulus of a different sensory modality. Taken together, our results suggest that, while there are differences in the processing of intervals marked by auditory and visual stimuli, they also share a common neural representation.

Heteroassociative storage of hippocampal pattern sequences in the CA3 subregion.

BACKGROUND: Recent research suggests that the CA3 subregion of the hippocampus has properties of both autoassociative network, due to its ability to complete partial cues, tolerate noise, and store associations between memories, and heteroassociative one, due to its ability to store and retrieve sequences of patterns. Although there are several computational models of the CA3 as an autoassociative network, more detailed evaluations of its heteroassociative properties are missing. METHODS: We developed a model of the CA3 subregion containing 10,000 integrate-and-fire neurons with both recurrent excitatory and inhibitory connections, and which exhibits coupled oscillations in the gamma and theta ranges. We stored thousands of pattern sequences using a heteroassociative learning rule with competitive synaptic scaling. RESULTS: We showed that a purely heteroassociative network model can (i) retrieve pattern sequences from partial cues with external noise and incomplete connectivity, (ii) achieve homeostasis regarding the number of connections per neuron when many patterns are stored when using synaptic scaling, (iii) continuously update the set of retrievable patterns, guaranteeing that the last stored patterns can be retrieved and older ones can be forgotten. DISCUSSION: Heteroassociative networks with synaptic scaling rules seem sufficient to achieve many desirable features regarding connectivity homeostasis, pattern sequence retrieval, noise tolerance and updating of the set of retrievable patterns.

Traumatic Ventricular Septal Defect After Stab Wound to the Chest Missed by Transthoracic Echocardiography: A Case Report.

Traumatic ventral septal defect may be sustained after either blunt force or penetrating trauma to the chest. Severity ranges from asymptomatic to acute decompensated heart failure. Our patient suffered a stab wound to the chest and was initially taken to the operating room for repair of a lacerated right ventricle. Subsequent postoperative hemodynamic deterioration prompted a bedside transthoracic echocardiogram, which failed to identify causal factors. A transesophageal echocardiogram performed immediately after ventral septal defect was demonstrated. This case serves to highlight the gaps in current standard practice and encourages the use of transesophageal echocardiogram as a screening tool in patients after penetrating cardiac injuries.

Dynamic representation of time in brain states.

The ability to process time on the scale of milliseconds and seconds is essential for behaviour. A growing number of studies have started to focus on brain dynamics as a mechanism for temporal encoding. Although there is growing evidence in favour of this view from computational and in vitro studies, there is still a lack of results from experiments in humans. We show that high-dimensional brain states revealed by multivariate pattern analysis of human EEG are correlated to temporal judgements. First, we show that, as participants estimate temporal intervals, the spatiotemporal dynamics of their brain activity are consistent across trials. Second, we present evidence that these dynamics exhibit properties of temporal perception, such as scale invariance. Lastly, we show that it is possible to predict temporal judgements based on brain states. These results show how scalp recordings can reveal the spatiotemporal dynamics of human brain activity related to temporal processing.

Can we use the 6-minute step test instead of the 6-minute walking test? An observational study.

OBJECTIVE: To verify whether or not heart rate is maintained below the calculated submaximal level in healthy, sedentary subjects when they perform the 6-minute step test (6MST) and the 6-minute walking test (6MWT), and to compare the maximal heart rate achieved by the subjects at the end of each test. DESIGN: Observational, cross-sectional study. SETTING: One tertiary centre. PARTICIPANTS: Two hundred and fifty-three participants from a pool of 330 healthy and sedentary subjects between 20 and 80 years of age. INTERVENTIONS: Both the 6MWT and the 6MST were performed in accordance with the American Thoracic Society's statement. Dyspnoea, blood pressure, oxygen saturation and heart rate were measured before and after each test. RESULTS: Mean heart rate immediately after the 6MST was significantly higher than mean heart rate immediately after the 6MWT {125 [standard deviation (SD) 19] vs 111 (SD 17) beats/minute; mean difference 13 (95% confidence interval of the difference 10 to 16); P<0.001}. Moreover, mean heart rate during (3minutes after commencement) the 6MST [118 (SD 18) beats/minute] was statistically higher than mean heart rate at the end of the 6MWT [111 (SD 18) beats/minute; P<0.001]. None of the subjects achieved the calculated submaximal heart rate. CONCLUSIONS: The 6MST and 6MWT are safe and produce submaximal effort in healthy participants. However, they are not interchangeable, and the 6MST requires more energy than the 6MWT.

Altered vesicular glutamate transporter distributions in the mouse cochlear nucleus following cochlear insult.

Vesicular glutamate transporters 1 and 2 (VGLUT1 and VGLUT2) have distinct distributions in the cochlear nucleus that correspond to sources of the labeled terminals. VGLUT1 is mainly associated with terminals of auditory nerve fibers, whereas VGLUT2 is mainly associated with glutamatergic terminals deriving from other sources that project to the cochlear nucleus (CN), including somatosensory and vestibular terminals. Previous studies in guinea pig have shown that cochlear damage results in a decrease of VGLUT1-labeled puncta and an increase in VGLUT2-labeled puncta. This indicates cross-modal compensation that is of potential importance in somatic tinnitus. To examine whether this effect is consistent across species and to provide a background for future studies, using transgenesis, the current study examines VGLUT expression profiles upon cochlear insult by intracochlear kanamycin injections in the mouse. Intracochlear kanamycin injections abolished ipsilateral ABR responses in all animals and reduced ipsilateral spiral ganglion neuron densities in animals that were sacrificed after four weeks, but not in animals that were sacrificed after three weeks. In all unilaterally deafened animals, VGLUT1 density was decreased in CN regions that receive auditory nerve fiber terminals, i.e., in the deep layer of the dorsal cochlear nucleus (DCN), in the interstitial region where the auditory nerve enters the CN, and in the magnocellular region of the antero- and posteroventral CN. In contrast, density of VGLUT2 expression was upregulated in the fusiform cell layer of the DCN and in the granule cell lamina, which are known to receive somatosensory and vestibular terminals. These results show that a cochlear insult induces cross-modal compensation in the cochlear nucleus of the mouse, confirming previous findings in guinea pig, and that these changes are not dependent on the occurrence of spiral ganglion neuron degeneration.