Á Szlávecz

6505809017

Publications - 16

NeuralODE-based Parameter Identification of the Three Chamber Model of the Circulatory System

Balázs Benyó Ákos Antal Bálint Szabó K. Kovács J. Geoffrey Chase Á Szlávecz

Publication Name: Iccc 2025 IEEE 12th International Joint Conference on Cybernetics and Computational Cybernetics Cyber Medical Systems Proceedings

Publication Date: 2025-01-01

Volume: Unknown

Issue: Unknown

Page Range: 161-166

Description:

In cases of Acute Circulatory Failure, fluid therapy is a commonly used intervention to stabilize heart function. However, the effectiveness of fluid therapy is not directly predictable, and the therapy can also be harmful. Physiological models can be used to predict fluid responsiveness - describing the effectiveness of fluid therapy for the patient - but require solving complex parameter identification problems. The current study aims to develop a Physics Informed Neural Network, specifically a NeuralODE-based parameter identification method for the Three Chamber cardiovascular model, which has the potential to be used to define a novel perfusion marker. The method is developed and validated on a clinical data set collected in model animal experiments.

Open Access: Yes

DOI: 10.1109/ICCC64928.2025.10999147

Comparison of three artificial intelligence methods for predicting 90% quantile interval of future insulin sensitivity of intensive care patients

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Omer S. Alkhafaf Ameer B. Alsultani Á Szlávecz B. Paláncz

Publication Name: IFAC Journal of Systems and Control

Publication Date: 2024-12-01

Volume: 30

Issue: Unknown

Page Range: Unknown

Description:

Insulin dosing of hyperglycemic patients in the intensive care unit (ICU) is a complex and nonlinear clinical control problem. Recent model-based glycemic control protocols predict a patient-specific and time-specific future insulin sensitivity distribution, which defines the future patient state in response to insulin and nutrition inputs. The prediction methods provide a 90% confidence interval for a future insulin sensitivity distribution for a given time horizon, making the prediction problem more specific compared to common prediction problems where the aim is to predict the expected value of the given stochastic parameter. This study proposes three alternative artificial intelligence-based insulin sensitivity prediction methods to improve the prediction accuracy and make prediction parameters better fit the clinical requirements. The proposed prediction methods use different neural network models: a classification deep neural network model, a Mixture Density Network model, and a Quantile Regression-based model. A large patient data set was used to create the neural network models, including 2357 patients and 92646 blood glucose measurements from three clinical sites (Christchurch, New Zealand, Gyula, Hungary, and Liege, Belgium). Prediction accuracy was assessed by statistical metrics expressing clinical requirements, as well as via validated in-silico virtual patient simulations comparing the clinical performance of a proven glycaemic control protocol using the alternative prediction methods to assess impact on glycemic control performance and thus the need for these alternative models.

Open Access: Yes

DOI: 10.1016/j.ifacsc.2024.100284

In-Silico Validation of Insulin Sensitivity Prediction by Neural Network-based Quantile Regression

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Omer S. Alkhafaf Ameer B. Alsultani Alaa N. Roel Petra Pintár Á Szlávecz B. Paláncz

Publication Name: IFAC Papersonline

Publication Date: 2024-09-01

Volume: 58

Issue: 24

Page Range: 368-373

Description:

High blood glucose levels and stress-induced hyperglycemia are common issues in intensive care units (ICU). Controlling blood glucose levels is crucial but challenging due to patient-specific variability. This challenge was addressed by developing model-based control protocols, which rely on identifying and predicting the patient-specific metabolic state. This study presents the in-silico simulation-based evaluation of a new artificial neural network-based insulin sensitivity (SI) prediction method. The models were trained on a dataset collected during clinical treatment using the stochastic-targeted (STAR) protocol and evaluated by simulating the clinical interventions on virtual patients created from retrospective clinical data. The results show the new models could be safely applied for SI prediction. Furthermore, the adopted method had very similar accuracy in the overall comparison of cohorts, with only minor differences noted in hypoglycemia events.

Open Access: Yes

DOI: 10.1016/j.ifacol.2024.11.065

Cardiovascular Model Identification Using Neural ODE

Balázs Benyó Ákos Antal J. Geoffrey Chase Bálint Szabó K. Kovács Cong Zhou Liam Murphy James Cushway Nicolas Davey Á Szlávecz B. Paláncz

Publication Name: IFAC Papersonline

Publication Date: 2024-09-01

Volume: 58

Issue: 24

Page Range: 374-379

Description:

Acute circulatory failure (ACF) is a clinical syndrome when the heart and circulatory circulation cannot provide adequate blood supply to meet metabolic needs of the organs. ACF affects 30%- 50% of intensive care unit (ICU) patients. Fluid resuscitation is the primary treatment of ACF. However, it fails in a significant proportion (about 50%) of cases due to lack of clinically feasible non-invasive perfusion markers to assess the efficacy of the fluid therapy. Unfortunately, unsuccessful fluid therapy negatively affects patient outcome, increasing ICU length of stay and costs. Recent studies show identifying Stressed Blood Volume (SBV) of the cardiovascular system can be used to assess the potential efficacy of fluid therapy. The development of the diagnostic method requires the identification of the central arterial pressure curve based on the femoral arterial pressure, which is clinically available. This central arterial pressure curve can be used to identify the cardiovascular system parameters. In this study, the main goal was to develop a parameter-identification method for the Tube-load model-based transfer function connecting the femoral and central arterial pressure curve by using the so-called Physics-informed Neural Network methodology, namely the Neural ODE method. The study presents the adaptation of the Neural ODE method to the given parameter identification problem and the validation of the developed identification method. The robustness of the developed identification method was tested and used on a series of measurement data recorded in animal experiments.

Open Access: Yes

DOI: 10.1016/j.ifacol.2024.11.066

Applying NeuralODE-based Cardiovascular Model Identification for Experimental Data Analysis

Balázs Benyó Ákos Antal J. Geoffrey Chase Bálint Szabó K. Kovács Á Szlávecz

Publication Name: Saci 2024 18th IEEE International Symposium on Applied Computational Intelligence and Informatics Proceedings

Publication Date: 2024-01-01

Volume: Unknown

Issue: Unknown

Page Range: 437-442

Description:

Recent model-based diagnostic methods have been found promising to provide non-invasive perfusion markers to assess the efficacy of fluid therapy, the most common treatment method for acute circulatory failure (ACF). The development of these model-based diagnostic methods requires the identification of the central arterial pressure curve based on the femoral arterial pressure. This current study presents improvements of the previously suggested NeuralODE-based identification method by suggesting the use of a physiologically interpretable parameter set of the Tube-load model-based transfer function for the physiological system analysis and suggesting a calculation method decreasing the measurement error-caused uncertainty of the identification parameter, called pulse transfer time.

Open Access: Yes

DOI: 10.1109/SACI60582.2024.10619737

Classification-based deep neural network vs mixture density network models for insulin sensitivity prediction problem

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Á Szlávecz B. Paláncz

Publication Name: Computer Methods and Programs in Biomedicine

Publication Date: 2023-10-01

Volume: 240

Issue: Unknown

Page Range: Unknown

Description:

Model-based glycemic control (GC) protocols are used to treat stress-induced hyperglycaemia in intensive care units (ICUs). The STAR (Stochastic-TARgeted) glycemic control protocol – used in clinical practice in several ICUs in New Zealand, Hungary, Belgium, and Malaysia – is a model-based GC protocol using a patient-specific, model-based insulin sensitivity to describe the patient's actual state. Two neural network based methods are defined in this study to predict the patient's insulin sensitivity parameter: a classification deep neural network and a Mixture Density Network based method. Treatment data from three different patient cohorts are used to train the network models. Accuracy of neural network predictions are compared with the current model- based predictions used to guide care. The prediction accuracy was found to be the same or better than the reference. The authors suggest that these methods may be a promising alternative in model-based clinical treatment for patient state prediction. Still, more research is needed to validate these findings, including in-silico simulations and clinical validation trials.

Open Access: Yes

DOI: 10.1016/j.cmpb.2023.107633

Comparison of Three Artificial Intelligence Methods for Predicting 90% Quantile Interval of Future Insulin Sensitivity of Intensive Care Patients

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Á Szlávecz B. Paláncz

Publication Name: IFAC Papersonline

Publication Date: 2023-07-01

Volume: 56

Issue: 2

Page Range: 2091-2095

Description:

Three alternative artificial intelligence-based insulin sensitivity prediction methods are compared in this study. Insulin sensitivity prediction is an essential step in calculating the optimal treatment options in model-based glycemic control protocol of insulin-dependent intensive care patients. The prediction methods must predict not only the expected value of the insulin sensitivity for a given time horizon but also the 90% confidence interval making the prediction problem more specific compared to the common prediction problems. All of the proposed prediction methods - proposed in our previous publications - use different neural network models: a classification deep neural network model, a Mixture Density Network based model, and a Quantile regression based model. The patent data set used for the development and accuracy assessment is from 3 clinical ICU cohorts, including 820 treatment episodes of 606 patients and 68,631 hours of treatment. To evaluate the efficacy of the prediction in the context of clinical requirements, three metrics are used Success rate, Interval ratio, and I-Score are applied.

Open Access: Yes

DOI: 10.1016/j.ifacol.2023.10.1110

Modeling the Correlation of Human Vertebral Body Volumes

Balázs Benyó P. Somogyi Bálint Szabó K. Kovács Sándor Bazsó Á Szlávecz Árpád Viola

Publication Name: IFAC Papersonline

Publication Date: 2023-07-01

Volume: 56

Issue: 2

Page Range: 9030-9035

Description:

Anatomical parameters of the human body strongly correlate with each other. Modelling these dependencies enables the creation of a realistic anatomical human body model that can be parameterized. Such a model can be used for several diagnostic processes to identify abnormalities or even give guidance in surgical interventions. This paper proposes a probabilistic model describing the dependencies between the vertebral body volumes of humans from the Caucasian human race. As demonstrated, the proposed model can accurately describe the relationship between the vertebral body volumes and is used for the prediction of an unknown vertebral volume based on a known one. The probabilistic model is created by using the CT segmentation of 37 patients.

Open Access: Yes

DOI: 10.1016/j.ifacol.2023.10.133

Increasing Patient Specificity of the Recurrent Neural Network Based Insulin Sensitivity Prediction by Transfer Learning

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Á Szlávecz B. Paláncz

Publication Name: Ines 2022 26th IEEE International Conference on Intelligent Engineering Systems 2022 Proceedings

Publication Date: 2022-01-01

Volume: Unknown

Issue: Unknown

Page Range: 27-32

Description:

Insulin therapy is a frequently applied treatment in intensive care to normalize the patient's blood glucose level increased by stress-induced hyperglycaemia. This therapy is generally referred to as Tight Glycaemic Control (TGC). The STAR (Stochastic-TARgeted) protocol is a TGC which uses the patient's insulin sensitivity (SI) as a key parameter to describe the patient's actual state. Prediction of the future patient's state, i.e. prediction of the patient's future SI value, is a crucial step of the protocol currently implemented by using the so-called Intensive Care INsulin Glucose (ICING) model of the human glucose-insulin system and an associated stochastic model. In our previous studies, we have shown that the Recurrent Neural Network (RNN) models are efficient alternative methods of SI prediction. In this paper, we suggest applying the so-called transfer learning technique to further enhance the accuracy of the SI prediction by using the SI history of the current patient. The paper presents the proposed methodology for applying transfer learning in SI prediction and the evaluation of the method's accuracy by comparing the outcomes with the currently applied solution. Insilico validation using real patients' data is involved in this validation.

Open Access: Yes

DOI: 10.1109/INES56734.2022.9922645

Artificial Intelligence Based Insulin Sensitivity Prediction for Personalized Glycaemic Control in Intensive Care

Balázs Benyó J. Geoffrey Chase Bálint Szabó K. Kovács Yahia Anane Á Szlávecz B. Paláncz

Publication Name: IFAC Papersonline

Publication Date: 2020-01-01

Volume: 53

Issue: 2

Page Range: 16335-16340

Description:

Stress-induced hyperglycaemia is a frequent complication in the intensive therapy that can be safely and efficiently treated by using the recently developed model-based tight glycaemic control (TGC) protocols. The most widely applied TGC protocol is the STAR (Stochastic-TARgeted) protocol which uses the insulin sensitivity (SI) for the assessment of the patients state. The patient-specific metabolic variability is managed by the so-called stochastic model allowing the prediction of the 90% confidence interval of the future SI value of the patients. In this paper deep neural network (DNN) based methods (classification DNN and Mixture Density Network) are suggested to implement the patient state prediction. The deep neural networks are trained by using three years of STAR treatment data. The methods are validated by comparing the prediction statistics with the reference data set. The prediction accuracy was also compared with the stochastic model currently used in the clinical practice. The presented results proved the applicability of the neural network based methods for the patient state prediction in the model based clinical treatment. Results suggest that the methods' prediction accuracy was the same or better than the currently used stochastic model. These results are the initial successful step in the validation process of the proposed methods which will be followed by in-silico simulation trials.

Open Access: Yes

DOI: 10.1016/j.ifacol.2020.12.659

Finite Element Simulation Based Analysis of Valve-sparing Aortic Root Surgery

Tamás Umenhoffer Balázs Benyó P. Somogyi K. Kovács Róbert Nagy Bernhard Laufer Anikó Kubovje Á Szlávecz Tamás Szerafin

Publication Name: IFAC Papersonline

Publication Date: 2020-01-01

Volume: 53

Issue: 2

Page Range: 16037-16042

Description:

The valve-sparing aortic root surgery is frequently used in the treatment of aortic root enlargement or aortic root aneurysm. The currently used common surgical practice assumes that the valve leaflets are distributed evenly around the circle defined by the aorta wall which is frequently a false assumption according to hart anatomy studies. A finite element simulation based method is proposed in this study for the analysis of the alternative surgical outcomes of the valve-sparing aortic root surgery. The simulation methods allow the definition of the aortic valve leaflet commissure positions and the diameter of the graft used to replace the aortic root. The suggested methods are able to estimate and quantitatively compare the hemodynamic functions and the robustness of the aortic valve functions. The corresponding modeling environment makes possible the easy definition of the patient specific aortic root model that is used as an input of the simulation. The initial validation of the simulation method was done by a real patient data based simulation study. These results suggest that the currently used surgical practice can be improved.

Open Access: Yes

DOI: 10.1016/j.ifacol.2020.12.409

Extension of a Glycaemic Control Medical Application with New Functions and Ergonomic User Interface Elements

Mihaela Lascu Balázs Benyó K. Kovács Brigitte Beatrix Schweisthal Á Szlávecz

Publication Name: 2018 13th International Symposium on Electronics and Telecommunications Isetc 2018 Conference Proceedings

Publication Date: 2018-12-19

Volume: Unknown

Issue: Unknown

Page Range: Unknown

Description:

The human body is composed of numerous highly complex metabolic processes. Abnormal variation of the blood glucose concentration is a common complication in the intensive care units. This paper presents a model-based method of glycaemic control and a medical application implementing the glycaemic control protocol. The principal cases were to extend the functionality of the application in order to automatically create multiple episodes for a patient already in the database of the application and an accessibly view for previous treatments calculation. These functions are very helpful and convenient because allow to the clinicians to have new measurements and a treatment calculation without entering the data of the patient again in the system. The visualization of previous treatments can be efficiently used to define optimized dosing for the subsequent treatment period. The paper concludes with the results of the extension of the medical application.

Open Access: Yes

DOI: 10.1109/ISETC.2018.8583885

Initial value selection of the model parameters in the curve fitting phase of the dynamic SPECT imaging

Balázs Benyó P. Somogyi G. Hesz K. Kovács Zsófia Barna Á Szlávecz

Publication Name: IFAC Papersonline

Publication Date: 2018-01-01

Volume: 51

Issue: 27

Page Range: 241-246

Description:

The dynamic SPECT (Single Photon Emission Computed Tomography) reconstruction algorithm developed in our prior work reconstructs the parameters of the time activity curve for each image voxel directly from the projection images. In each iterations of the SPECT reconstruction beyond the static 3D MLEM (Maximum Likelihood Expectation Maximization) step, the algorithm performs a fitting process for each voxel in order to estimate the parameters describing the function of the examined organ considering that the time frames are not independent from each other. In real cases the fitted curve is nonlinear function of these parameters, it is usually described as the sum of exponential functions. In order to estimate the parameters properly, an iterative root-finding method is applied. In the current study the Newton-Raphson method is used. The selection of a proper initial value for the root-finding method is critical in order to achieve convergence of the fitting process. If the initial guess is not appropriate, the root-finding algorithm can diverge or converge to an inappropriate parameter set that can result in unacceptable reconstructed parameters. This affects then the subsequent MLEM iterations, also neighboring voxels and breaks the reconstruction. In this work we investigated different methods to calculate the initial values of the fitting process and evaluated the reconstructed parameter set of the dynamic SPECT reconstruction algorithm. Three different methods are investigated, one that uses the fitted parameters of the previous MLEM iteration, one that is based on the sum of the geometrical series of the exponentials and one that calculates the best guess using both methods. The three methods were compared by benchmark reconstruction cases using a mathematical phantom. In each reconstruction different initial value selection method was applied then the time activity curves of the voxels belonging to the same tissue were statistically evaluated using the reconstructed parameters. In the study no significant differences were found in the mean value of the reconstructed parameters. The standard deviation of the parameters was similar between the two simple approaches, however, the combination of the methods resulted in better statistical performance.

Open Access: Yes

DOI: 10.1016/j.ifacol.2018.11.635

Investigation of 3D SPECT reconstruction with multi-energy photon emitters 1

Balázs Benyó G. Hesz K. Kovács Á Szlávecz

Publication Name: IFAC Papersonline

Publication Date: 2015-09-01

Volume: 28

Issue: 20

Page Range: 30-35

Description:

Parallel projection based Single Emission Computed Tomography is a widely used nuclear imaging technique. Non-homogeneous attenuation medium and the distance dependent spacial resolution (DDSR) of the parallel imaging cause serious artefacts during image reconstruction. Both effects are dependent on the energy of gamma photons used for imaging. In this paper an efficient parallel reconstruction algorithm is introduced that is executed on the Graphics Processing Unit. The aim of the presented study was to investigate the possibilities of reconstruction techniques when multi-energy photon emitters are used. An analytical 3D projector with attenuation and detector response modelling was used to generate projection sets for Gallium-67 isotope studies. Data were reconstructed from one photopeak only using the corresponding attenuation map. The projection data sets were added together and reconstructed using average attenuation and DDSR compensation values as well as simulating every photopeak individually. The third combined reconstruction technique was using every projection set as separate measurement. In this study we showed that all three strategies result in similar image quality, however the average attenuation correction method is computationally less demanding.

Open Access: Yes

DOI: 10.1016/j.ifacol.2015.10.110

Reconstruction of myocardial short-scan SPECT images

Balázs Benyó Á Szlávecz Péter Várady Tamás Bükki

Publication Name: Ines 05 IEEE 9th International Conference on Intelligent Engineering Systems Proceedings

Publication Date: 2005-01-01

Volume: 2005

Issue: Unknown

Page Range: 37-41

Description:

In myocardial perfusion SPECT imaging the effect of photon attenuation may introduce artifacts in the reconstructed image due to the highly non-uniform distribution of tissue in the thorax region, potentially resulting in false-positive interpretations. It was the general consideration that the adequate compensation of photon attenuation requires, that the emission data be measured at projection angles over 2π in the case of the attenuation medium is inhomogeneous. The reduction of the scanning angle in SPECT imaging may be desirable because it can reduce scanning time and thereby minimize patient-motion and other artifacts. In SPECT myocardial imaging emission data is measured historically at projection angles over π from the right anterior oblique (RAO) to the left posterior oblique (LPO). This configuration results in better image contrast and, in some cases, betters spatial resolution. However, in this case the reconstructed image may suffer more severely from geometric distortion than 2π angular sampling. It has been proven recently in analytical computer simulation studies that the data function over 2π in SPECT with non-uniform attenuation contains redundant information; therefore the scanning angle theoretically can be reduced from 2π to π without loss of information. In this study our goal was to investigate how the various short-scan SPECT scheme configurations works in a real myocardial SPECT imaging system with highly inhomogeneous attenuating medium using attenuation correction. The measured projection images were reconstructed using the Maximum Likelihood Expectation Maximization algorithm with attenuation correction. The reconstructed slices of the various short-scan configurations and the full-scan slices were compared by a cardiac stress/rest software package. © 2005 IEEE.

Open Access: Yes

DOI: 10.1109/INES.2005.1555127

A common periodic table of codons and amino acids

Balázs Benyó G. Fördös J. C. Biro Á Szlávecz T. Micsik Zoltán Benyó C. Sansom

Publication Name: Biochemical and Biophysical Research Communications

Publication Date: 2003-06-27

Volume: 306

Issue: 2

Page Range: 408-415

Description:

A periodic table of codons has been designed where the codons are in regular locations. The table has four fields (16 places in each) one with each of the four nucleotides (A, U, G, C) in the central codon position. Thus, AAA (lysine), UUU (phenylalanine), GGG (glycine), and CCC (proline) were placed into the corners of the fields as the main codons (and amino acids) of the fields. They were connected to each other by six axes. The resulting nucleic acid periodic table showed perfect axial symmetry for codons. The corresponding amino acid table also displaced periodicity regarding the biochemical properties (charge and hydropathy) of the 20 amino acids and the position of the stop signals. The table emphasizes the importance of the central nucleotide in the codons and predicts that purines control the charge while pyrimidines determine the polarity of the amino acids. This prediction was experimentally tested. © 2003 Elsevier Science (USA). All rights reserved.

Open Access: Yes

DOI: 10.1016/S0006-291X(03)00974-4