Prediction of Intracranial Pressure using Standard Periodic Vital Signs

[miticdalibor]

Author

• Name: Dalibor Mitic

• Affiliation: FH Kufstein

Keywords

• Artificial Intelligence
• intracranial pressure
• medical health
• neurosurgery
• Machine Learning
• prediction
• vital signs
• traumatic brain injury

Abstract
Currently, to measure the intracranial pressure (ICP) there are invasive probes, which need to be positioned intraoperatively during primary surgery or as an additional bed-side procedure through a standardized frontal approach targeting the lateral ventricular system or the white matter of the non-dominant or predominantly lesioned hemisphere. As this intraoperatively approach is linked with risks, the medical research aims to find potential correlators for the ICP by investigating different biomarkers. This paper proposes a machine learning approach to predict the ICP using standard periodic vital signs and identifying potential correlators by investigating the feature importance of the models instead of using basic statistics.

Full PDF: DOI, [Zenodo],
Github Repo.
Github Release
Zenodo WholeTale
Zenodo Github
..
Prediction_of_Intracranial_Pressure_using_Standard_Periodic_Vital_Signs.pdf

[MHeini]

This article is an excellent and informative research article that proposes a machine learning approach to predict intracranial pressure (ICP) using standard periodic vital signs and to identify potential correlates by examining the significance of features in the models. The invasive methods currently used to measure ICP and the associated risks are highlighted, and the potential benefits of a machine learning approach are well justified. The results of the study show that the models are able to predict intracranial pressure using regular vital signs, and that the inclusion of the acute psychiatric score as a feature could improve the prediction of the model. In the "Outlook" section, promising future research directions are suggested, such as the use of deep neural networks to further improve prediction and the study of feature importance using the feature permutation approach. Also noteworthy is the idea of including a data set with frequent calculations of the acute physiology score if continuous measurement of vital signs is difficult to achieve. Overall, this article represents a valuable contribution to medical research and has the potential to improve patient outcomes.

[kayagoekan]

Order accepted for review

[renatofnc]

Prediction of intracranial pressure using standard periodic vital signs
Author: Dalibor Mitic
Reviewer: Renato Fonseca Nobrega
With his paper, Dalibor Mitic addresses one possible alternative regarding the measurement of the intracranial pressure (ICP) under assistance of machine learning techniques. For this purpose, he gives a very clear introduction about the invasive nature of ICP, correctly mentioning the risks of this approach. The alternative offered by the author builds on using standard periodic vital signs as predictors for detecting ICP levels. The author formulates a clear research question based on these vitals and builds an interesting analytic experiment to investigate this question on an explorative manner by himself. Before initiating the analysis, the author takes the time to explain the origin and meaning of the data thoroughly. He also explicitly offers a plausible explanation for some inconsistencies regarding the raw data. The present work is clearly oriented along the phases of CRISP-ML, and based on that, it exposes initially concise thoughts about the steps to be taken regarding data preprocessing. The author suggests sharp actions to address the current state of some erroneous data observations, achieving good results at the end of this stage. The following model development take place in a well-structured way, broken down into two phases of creating a baseline model and improving its results through further iterations. Here I would have wished to know, why XGBoost was the algorithm of the choice (maybe performance thoughts) and if and how a hyperparameter tuning has taken place (I suppose yes). Nevertheless, the author is able to achieve improvements in the following iterations and closes his experiment making the plausible conclusion, that vital signs can offer a good approximation for ICP values. Mitic can cast an outlook for further possibilities, closing a well-organized article competently.

One minor tweak could be taken: in section 3.3 the table apachePatientResult is probably the one meant to be described, not the vitalPeriodics.

Points(20/20)

[kayagoekan]

This scientific article deals intensively with the measurement of intracranial pressure (ICP) with the use of machine learning techniques. Regarding this topic, the author of this article gives a very good introduction about the invasive nature of ICP. Furthermore, the risks of this approach are discussed.
For the application proposed here by Dalibor Mitic was the use of standard periodic vital signs as predictors for the detection of ICP values. In this informative research article, the potential benefits of a machine learning approach were very well justified. Thus, this article clearly demonstrates that models are able to predict intracranial pressure using regular vital signs. Furthermore, results of this study show that the inclusion of the acute psychiatric score as a sign would improve the prediction of the model. The author hereby suggests that the use of neurological networks can be used to improve prediction. This article is a valuable contribution to the field and with further research would lead to solid results in science. The article is very well structured and the language is very easy to understand.

Points (20/20)