Gated Transformer for Decoding Human Brain EEG Signals.

In this work, we propose to use a deep learning framework for decoding the electroencephalogram (EEG) signals of human brain activities. More specifically, we learn an end-to-end model that recognizes natural images or motor imagery by the EEG data that is collected from the corresponding human neural activities. In order to capture the temporal information encoded in the long EEG sequences, we first employ an enhanced version of Transformer, i.e., gated Transformer, on EEG signals to learn the feature representation along a sequence of embeddings. Then a fully-connected Softmax layer is used to predict the classification results of the decoded representations. To demonstrate the effectiveness of the gated Transformer approach, we conduct experiments on the image classification task for a human brain-visual dataset and the classification task for a motor imagery dataset. The experimental results show that our method achieves new state-of-the-art performance compared to multiple existing methods that are widely used for EEG classification.

An early respiratory distress detection method with Markov models.

A method for early detection of respiratory distress in hospitalized patients which is based on a multi-parametric analysis of respiration rate (RR) and pulse oximetry (SpO2) data trends to ascertain patterns of patient instability pertaining to respiratory distress is described. Current practices of triggering caregiver alerts are based on simple numeric threshold breaches of SpO2. The pathophysiological patterns of respiratory distress leading to in-hospital deaths are much more complex to be detected by numeric thresholds. Our pattern detection algorithm is based on a Markov model framework based on multi-parameter pathophysiological patterns of respiratory distress, and triggers in a timely manner and prior to the violation of SpO2 85-90% threshold, providing additional lead time to attempt to reverse the deteriorating state of the patient. We present the performance of the algorithm on MIMIC II dataset resulting in true positive rate of 92% and false positive rate of 6%.

Prediction of mortality from respiratory distress among long-term mechanically ventilated patients.

With the advent of inexpensive storage, pervasive networking, and wireless devices, it is now possible to store a large proportion of the medical data that is collected in the intensive care unit (ICU). These data sets can be used as valuable resources for developing and validating predictive analytics. In this report, we focus on the problem of prediction of mortality from respiratory distress among long-term mechanically ventilated patients using data from the publicly-available MIMIC-II database. Rather than only reporting p-values for univariate or multivariate regression, as in previous work, we seek to generate sparsest possible model that will predict mortality. We find that the presence of severe sepsis is highly associated with mortality. We also find that variables related to respiration rate have more predictive accuracy than variables related to oxygenation status. Ultimately, we have developed a model which predicts mortality from respiratory distress in the ICU with a cross-validated area-under-the-curve (AUC) of approximately 0.74. Four methodologies are utilized for model dimensionality-reduction: univariate logistic regression, multivariate logistic regression, decision trees, and penalized logistic regression.

Hemodynamic-impact-based prioritization of ventricular tachycardia alarms.

Ventricular tachycardia (V-tach) is a very serious condition that occurs when the ventricles are driven at high rates. The abnormal excitation pathways make ventricular contraction less synchronous resulting in less effective filling and emptying of the left ventricles. However, almost half of the V-tach alarms declared through processing of patterns observed in electrocardiography are not clinically actionable. The focus of this study is to provide guidance on determining whether a technically-correct V-tach alarm is clinically-actionable by determining its "hemodynamic impact". A supervisory learning approach based on conditional inference trees to determine the hemodynamic impact of a V-tach alarm based on extracted features is described. According to preliminary results on a subset of Multiparameter intelligent monitoring in intensive care II (MIMIC-II) database, true positive rate of more than 90% can be achieved.

Prediction of mean arterial blood pressure with linear stochastic models.

A model-based approach that integrates known portion of the cardiovascular system and unknown portion through a parameter estimation to predict evolution of the mean arterial pressure is considered. The unknown portion corresponds to the neural portion that acts like a controller that takes corrective actions to regulate the arterial blood pressure at a constant level. The input to the neural part is the arterial pressure and output is the sympathetic nerve activity. In this model, heart rate is considered a proxy for sympathetic nerve activity. The neural portion is modeled as a linear discrete-time system with random coefficients. The performance of the model is tested on a case study of acute hypotensive episodes (AHEs) on PhysioNet data. TPRs and FPRs improve as more data becomes available during estimation period.