Abstract Background: Artificial intelligence (AI) techniques play a major role in anesthesiology, even though their importance is often overlooked. In the extant literature, AI approaches, such as artificial neural networks (ANNs), have been underutilized, being used mainly to model patient's consciousness state, to predict the precise number of anesthetic gases, the level of analgesia, or the need of anesthesiological blocks, among others. In the field of neurosurgery, ANNs have been effectively applied to the diagnosis and prognosis of cerebral tumors, seizures, low back pain, and also to the monitoring of intracranial pressure (ICP). Methods: A multilayer perceptron (MLP), which is a feedforward ANN, with hyperbolic tangent as activation function in the input/hidden layers, softmax as activation function in the output layer, and cross-entropy as error function, was used to model the impact of prone versus supine position and the use of positive end expiratory pressure (PEEP) on ICP in a sample of 30 patients undergoing spinal surgery. Different noninvasive surrogate estimations of ICP have been used and compared: namely, mean optic nerve sheath diameter (ONSD), noninvasive estimated cerebral perfusion pressure (NCPP), Pulsatility Index (PI), ICP derived from PI (ICP-PI), and flow velocity diastolic formula (FVDICP). Results: ONSD proved to be a more robust surrogate estimation of ICP, with a predictive power of 75%, whilst the power of NCPP, ICP-PI, PI, and FVDICP were 60.5%, 54.8%, 53.1%, and 47.7%, respectively. Conclusions: Our MLP analysis confirmed our findings previously obtained with regression, correlation, multivariate receiving operator curve (multi-ROC) analyses. ANNs can be successfully used to predict the effects of prone versus supine position and PEEP on ICP in patients undergoing spinal surgery using different noninvasive surrogate estimators of ICP.
Artificial neural networks can be effectively used to model changes of intracranial pressure (ICP) during spinal surgery using different non invasive ICP surrogate estimators / Watad, A; Bragazzi, Nl; Bacigaluppi, S; Amital, H; Watad, S; Sharif, K; Bisharat, B; Siri, A; Mahamid, A; Abu Ras, H; Nasr, A; Bilotta, F; Robba, C; Adawi, M. Artificial neural networks can be effectively used to model changes of intracranial pressure (ICP) during spinal surgery using different non invasive ICP surrogate estimators.. - In: JOURNAL OF NEUROSURGICAL SCIENCES. - ISSN 0390-5616. - (2018).
Artificial neural networks can be effectively used to model changes of intracranial pressure (ICP) during spinal surgery using different non invasive ICP surrogate estimators
Bilotta F;
2018
Abstract
Abstract Background: Artificial intelligence (AI) techniques play a major role in anesthesiology, even though their importance is often overlooked. In the extant literature, AI approaches, such as artificial neural networks (ANNs), have been underutilized, being used mainly to model patient's consciousness state, to predict the precise number of anesthetic gases, the level of analgesia, or the need of anesthesiological blocks, among others. In the field of neurosurgery, ANNs have been effectively applied to the diagnosis and prognosis of cerebral tumors, seizures, low back pain, and also to the monitoring of intracranial pressure (ICP). Methods: A multilayer perceptron (MLP), which is a feedforward ANN, with hyperbolic tangent as activation function in the input/hidden layers, softmax as activation function in the output layer, and cross-entropy as error function, was used to model the impact of prone versus supine position and the use of positive end expiratory pressure (PEEP) on ICP in a sample of 30 patients undergoing spinal surgery. Different noninvasive surrogate estimations of ICP have been used and compared: namely, mean optic nerve sheath diameter (ONSD), noninvasive estimated cerebral perfusion pressure (NCPP), Pulsatility Index (PI), ICP derived from PI (ICP-PI), and flow velocity diastolic formula (FVDICP). Results: ONSD proved to be a more robust surrogate estimation of ICP, with a predictive power of 75%, whilst the power of NCPP, ICP-PI, PI, and FVDICP were 60.5%, 54.8%, 53.1%, and 47.7%, respectively. Conclusions: Our MLP analysis confirmed our findings previously obtained with regression, correlation, multivariate receiving operator curve (multi-ROC) analyses. ANNs can be successfully used to predict the effects of prone versus supine position and PEEP on ICP in patients undergoing spinal surgery using different noninvasive surrogate estimators of ICP.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
