Topological data analysis (TDA) enhances bispectral EEG (BSEEG) algorithm for detection of delirium.

Current methods for screening and detecting delirium are not practical in clinical settings. We previously showed that a simplified EEG with bispectral electroencephalography (BSEEG) algorithm can detect delirium in elderly inpatients. In this study, we performed a post-hoc BSEEG data analysis using larger sample size and performed topological data analysis to improve the BSEEG method. Data from 274 subjects included in the previous study were analyzed as a 1st cohort. Subjects were enrolled at the University of Iowa Hospitals and Clinics (UIHC) between January 30, 2016, and October 30, 2017. A second cohort with 265 subjects was recruited between January 16, 2019, and August 19, 2019. The BSEEG score was calculated as a power ratio between low frequency to high frequency using our newly developed algorithm. Additionally, Topological data analysis (TDA) score was calculated by applying TDA to our EEG data. The BSEEG score and TDA score were compared between those patients with delirium and without delirium. Among the 274 subjects from the first cohort, 102 were categorized as delirious. Among the 206 subjects from the second cohort, 42 were categorized as delirious. The areas under the curve (AUCs) based on BSEEG score were 0.72 (1st cohort, Fp1-A1), 0.76 (1st cohort, Fp2-A2), and 0.67 (2nd cohort). AUCs from TDA were much higher at 0.82 (1st cohort, Fp1-A1), 0.84 (1st cohort, Fp2-A2), and 0.78 (2nd cohort). When sensitivity was set to be 0.80, the TDA drastically improved specificity to 0.66 (1st cohort, Fp1-A1), 0.72 (1st cohort, Fp2-A2), and 0.62 (2nd cohort), compared to 0.48 (1st cohort, Fp1-A1), 0.54 (1st cohort, Fp2-A2), and 0.46 (2nd cohort) with BSEEG. BSEEG has the potential to detect delirium, and TDA is helpful to improve the performance.

Evaluation of point-of-care thumb-size bispectral electroencephalography device to quantify delirium severity and predict mortality.

BACKGROUND: We have developed the bispectral electroencephalography (BSEEG) method for detection of delirium and prediction of poor outcomes. AIMS: To improve the BSEEG method by introducing a new EEG device. METHOD: In a prospective cohort study, EEG data were obtained and BSEEG scores were calculated. BSEEG scores were filtered on the basis of standard deviation (s.d.) values to exclude signals with high noise. Both non-filtered and s.d.-filtered BSEEG scores were analysed. BSEEG scores were compared with the results of three delirium screening scales: the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU), the Delirium Rating Scale-Revised-98 (DRS) and the Delirium Observation Screening Scale (DOSS). Additionally, the 365-day mortalities and the length of stay (LOS) in the hospital were analysed. RESULTS: We enrolled 279 elderly participants and obtained 620 BSEEG recordings; 142 participants were categorised as BSEEG-positive, reflecting slower EEG activity. BSEEG scores were higher in the CAM-ICU-positive group than in the CAM-ICU-negative group. There were significant correlations between BSEEG scores and scores on the DRS and the DOSS. The mortality rate of the BSEEG-positive group was significantly higher than that of the BSEEG-negative group. The LOS of the BSEEG-positive group was longer compared with that of the BSEEG-negative group. BSEEG scores after s.d. filtering showed stronger correlations with delirium screening scores and more significant prediction of mortality. CONCLUSIONS: We confirmed the usefulness of the BSEEG method for detection of delirium and of delirium severity, and prediction of patient outcomes with a new EEG device.

Personalized Perioperative Multi-scale, Multi-physics Heart Simulation of Double Outlet Right Ventricle.

For treatment of complex congenital heart disease, computer simulation using a three-dimensional heart model may help to improve outcomes by enabling detailed preoperative evaluations. However, no highly integrated model that accurately reproduces a patient's pathophysiology, which is required for this simulation has been reported. We modelled a case of complex congenital heart disease, double outlet right ventricle with ventricular septal defect and atrial septal defect. From preoperative computed tomography images, finite element meshes of the heart and torso were created, and cell model of cardiac electrophysiology and sarcomere dynamics was implemented. The parameter values of the heart model were adjusted to reproduce the patient's electrocardiogram and haemodynamics recorded preoperatively. Two options of in silico surgery were performed using this heart model, and the resulting changes in performance were examined. Preoperative and postoperative simulations showed good agreement with clinical records including haemodynamics and measured oxyhaemoglobin saturations. The use of a detailed sarcomere model also enabled comparison of energetic efficiency between the two surgical options. A novel in silico model of congenital heart disease that integrates molecular models of cardiac function successfully reproduces the observed pathophysiology. The simulation of postoperative state by in silico surgeries can help guide clinical decision-making.

Absence of Rapid Propagation through the Purkinje Network as a Potential Cause of Line Block in the Human Heart with Left Bundle Branch Block.

Background: Cardiac resynchronization therapy is an effective device therapy for heart failure patients with conduction block. However, a problem with this invasive technique is the nearly 30% of non-responders. A number of studies have reported a functional line of block of cardiac excitation propagation in responders. However, this can only be detected using non-contact endocardial mapping. Further, although the line of block is considered a sign of responders to therapy, the mechanism remains unclear. Methods: Herein, we created two patient-specific heart models with conduction block and simulated the propagation of excitation based on a cellmodel of electrophysiology. In one model with a relatively narrow QRS width (176 ms), we modeled the Purkinje network using a thin endocardial layer with rapid conduction. To reproduce a wider QRS complex (200 ms) in the second model, we eliminated the Purkinje network, and we simulated the endocardial mapping by solving the inverse problem according to the actual mapping system. Results: We successfully observed the line of block using non-contact mapping in the model without the rapid propagation of excitation through the Purkinje network, although the excitation in the wall propagated smoothly. This model of slow conduction also reproduced the characteristic properties of the line of block, including dense isochronal lines and fractionated local electrocardiograms. Further, simulation of ventricular pacing from the lateral wall shifted the location of the line of block. By contrast, in the model with the Purkinje network, propagation of excitation in the endocardial map faithfully followed the actual propagation in the wall, without showing the line of block. Finally, switching the mode of propagation between the two models completely reversed these findings. Conclusions: Our simulation data suggest that the absence of rapid propagation of excitation through the Purkinje network is the major cause of the functional line of block recorded by non-contact endocardial mapping. The line of block can be used to identify responders as these patients loose rapid propagation through the Purkinje network.

Multi-scale, tailor-made heart simulation can predict the effect of cardiac resynchronization therapy.

BACKGROUND: The currently proposed criteria for identifying patients who would benefit from cardiac resynchronization therapy (CRT) still need to be optimized. A multi-scale heart simulation capable of reproducing the electrophysiology and mechanics of a beating heart may help resolve this problem. The objective of this retrospective study was to test the capability of patient-specific simulation models to reproduce the response to CRT by applying the latest multi-scale heart simulation technology. METHODS AND RESULTS: We created patient-specific heart models with realistic three-dimensional morphology based on the clinical data recorded before treatment in nine patients with heart failure and conduction block treated by biventricular pacing. Each model was tailored to reproduce the surface electrocardiogram and hemodynamics of each patient in formats similar to those used in clinical practice, including electrocardiography (ECG), echocardiography, and hemodynamic measurements. We then performed CRT simulation on each heart model according to the actual pacing protocol and compared the results with the clinical data. CRT simulation improved the ECG index and diminished wall motion dyssynchrony in each patient. These results, however, did not correlate with the actual response. The best correlation was obtained between the maximum value of the time derivative of ventricular pressure (dP/dtmax) and the clinically observed improvement in the ejection fraction (EF) (r=0.94, p<0.01). CONCLUSIONS: By integrating the complex pathophysiology of the heart, patient-specific, multi-scale heart simulation could successfully reproduce the response to CRT. With further verification, this technique could be a useful tool in clinical decision making.

Patient specific simulation of body surface ECG using the finite element method.

BACKGROUND: Recent studies, supported by advances in computer science, have successfully simulated the excitation and repolarization processes of the heart, based on detailed cell models of electrophysiology and implemented with realistic morphology. METHODS: In this study, we extend these approaches to simulate the body surface electrocardiogram (ECG) of specific individuals. Patient-specific finite element models of the heart and torso are created for four patients with various heart diseases, based on clinical data including computer tomography, while the parallel multi-grid method is used to solve the dynamic bi-domain problem. Personalization procedures include demarcation of nonexcitable tissue, allocation of the failing myocyte model of electrophysiology, and modification of the excitation sequence. In particular, the adjustment of QRS morphology requires iterative computations, facilitated by the simultaneous visualization of the propagation of excitation in the heart, average QRS vector in the torso, and 12-lead ECG. RESULTS: In all four cases we obtained reasonable agreement between the simulated and actual ECGs. Furthermore, we also simulated the ECGs of three of the patients under bi-ventricular pacing, and once again successfully reproduced the actual ECG morphologies. Since no further adjustments were made to the heart models in the pacing simulations, the good agreement provides strong support for the validity of the models. CONCLUSIONS: These results not only help us understand the cellular basis of the body surface ECG, but also open the possibility of heart simulation for clinical applications.

Multi-scale computational modelling in biology and physiology.

Recent advances in biotechnology and the availability of ever more powerful computers have led to the formulation of increasingly complex models at all levels of biology. One of the main aims of systems biology is to couple these together to produce integrated models across multiple spatial scales and physical processes. In this review, we formulate a definition of multi-scale in terms of levels of biological organisation and describe the types of model that are found at each level. Key issues that arise in trying to formulate and solve multi-scale and multi-physics models are considered and examples of how these issues have been addressed are given for two of the more mature fields in computational biology: the molecular dynamics of ion channels and cardiac modelling. As even more complex models are developed over the coming few years, it will be necessary to develop new methods to model them (in particular in coupling across the interface between stochastic and deterministic processes) and new techniques will be required to compute their solutions efficiently on massively parallel computers. We outline how we envisage these developments occurring.