Work in the last two decades has shown that neural mass models (NMM) can realistically reproduce and explain epileptic seizure transitions as recorded by electrophysiological methods (EEG, SEEG). In previous work, advances were achieved by increasing excitation and heuristically varying network inhibitory coupling parameters in the models. Based on these early studies, we provide a laminar NMM capable of realistically reproducing the electrical activity recorded by SEEG in the epileptogenic zone during interictal to ictal states. With the exception of the external noise input into the pyramidal cell population, the model dynamics are autonomous. By setting the system at a point close to bifurcation, seizure-like transitions are generated, including pre-ictal spikes, low voltage fast activity, and ictal rhythmic activity. A novel element in the model is a physiologically motivated algorithm for chloride dynamics: the gain of GABAergic post-synaptic potentials is modulated by the pathological accumulation of chloride in pyramidal cells due to high inhibitory input and/or dysfunctional chloride transport. In addition, in order to simulate SEEG signals for comparison with real seizure recordings, the NMM is embedded first in a layered model of the neocortex and then in a realistic physical model. We compare modeling results with data from four epilepsy patient cases. By including key pathophysiological mechanisms, the proposed framework captures succinctly the electrophysiological phenomenology observed in ictal states, paving the way for robust personalization methods based on NMMs.

The prospect of personalized computational modeling in neurological disorders, and in particular in epilepsy, is poised to revolutionize the field. Work in the last two decades has demonstrated that neural mass models (NMM) can realistically reproduce and explain epileptic seizure transitions as recorded by electrophysiological methods (EEG, SEEG). In previous work, advances were achieved by i) increasing excitation in NMM and ii) heuristically varying network inhibitory coupling parameters or, equivalently, inhibitory synaptic gains. Based on those studies, we provide here a laminar neural mass model capable of realistically reproducing the electrical activity recorded by SEEG in the epileptogenic zone during interictal to ictal states. With the exception of the external noise input onto the pyramidal cell population, the model dynamics are autonomous --- all model parameters are static. By setting the system at a point close to bifurcation, seizure-like transitions are generated, including pre-ictal spikes, low voltage fast activity, and ictal rhythmic activity. A novel element in the model is a physiologically plausible algorithm for chloride accumulation dynamics: the gain of GABAergic post-synaptic potentials is modulated by the pathological accumulation of Cl$^-$ in pyramidal cells, due to high inhibitory input and/or dysfunctional chloride transport. In addition, in order to simulate SEEG signals to compare with real recordings performed in epileptic patients, the NMM is embedded first in a layered model of the neocortex and then in a realistic physical model. We compare modeling results with data from four epilepsy patient cases. By including key pathophysiological mechanisms, the proposed framework captures succinctly the electrophysiological phenomenology observed in ictal states, paving the way for robust personalization methods using brain network models based on NMMs.

No abstract

No abstract

Background and Aims Patients with Acute Renal Failure (ARF) have a high risk of mortality, especially those who enter the Intensive Care Unit (ICU). In this population, predictive models of mortality on prognostic scales, such as SAPS-II (Simplified Acute Physiology Score II), linearly relate risk factors without taking into account the complex relationship's variables can have. There are models where Machine Learning (ML) techniques have been used, but there is still room for improvement. The implementation of deep artificial neural networks (DANN) can be challenging. The literature models, using SAPS-II report an accuracy, f1 and ROC area (receiver operating curve) in ranges of 0.538-0.621, 0.333-0.377 and 0.781-0.809 respectively. The best results with ML are improved in neural networks of a hidden layer or random forest, being the best performance in the latter: accuracy 0.715-0.741, F1 0.449-0.470 and ROC between 0.862-0.870. The aim is to evaluate and improve the predictive capacity of ML techniques for the prediction of mortality in patients with ARF admitted to the ICU, through the use of the open database MIMIC-III (Medical Information Martfor Intensive Care III). Method Design: Retrospective analysis of historical cooperation of 20,928 patients with ARF from Beth Israel Deaconess (Boston), from 2001 to 2012. Method: ML algorithm based on DANN. Creation of a model to predict in-hospital mortality after discharge from the ICU with the variables of the first 24 hours after admission to the ICU. To evaluate the robustness of the model has been performed cross-validation by separating the samples into different combinations of training and test data (k-folds). The unavailable variables haematological with means extracted from the training set of the respective fold. The DANN trained with the complex relationship's variables folds, two hidden layers of 75 and 40 neurons respectively. Inclusion criteria: > 16 years, AKI according to KDIGO criteria and predictive variables based on SAPS-II. Variables: Age, sex, type of admission, number of admissions in ICU, heart rate, blood pressure, temperature, PaO2, FiO2, sodium, potassium, bilirubin, bicarbonate, urea, leukocytes, diuresis and diagnosis of ICD-9 (metastatic cancer, haematological malignancies). Results Accuracy: 76.7% 0 ± 1.7%, f1: 86.6% ± 1.9%, RO area: 0.859 ± 0.006, sensitivity: 75.7% ± 2.8%, specificity: 80.5% ± 2.9%. Conclusion The use of Machine Learning techniques based on deep artificial neural networks can improve the predictive ability of mortality in acute renal failure of the critically ill patient of traditional clinical risk scales and even current Machine Learning models.