A Methodology Based on Expert Systems for the Early Detection and Prevention of Hypoxemic Clinical Cases

Abstract

Respiratory diseases are currently considered to be amongst the most frequent causes of death and disability worldwide, and even more so during the year 2020 because of the COVID-19 global pandemic. Aiming to reduce the impact of these diseases, in this work a methodology is developed that allows the early detection and prevention of potential hypoxemic clinical cases in patients vulnerable to respiratory diseases. Starting from the methodology proposed by the authors in a previous work and grounded in the definition of a set of expert systems, the methodology can generate alerts about the patient’s hypoxemic status by means of the interpretation and combination of data coming both from physical measurements and from the considerations of health professionals. A concurrent set of Mamdani-type fuzzy-logic inference systems allows the collecting and processing of information, thus determining a final alert associated with the measurement of the global hypoxemic risk. This new methodology has been tested experimentally, producing positive results so far from the viewpoint of time reduction in the detection of a blood oxygen saturation deficit condition, thus implicitly improving the consequent treatment options and reducing the potential adverse effects on the patient’s health.

Design and Development of a New Methodology Based on Expert Systems Applied to the Prevention of Indoor Radon Gas Exposition Risks

“…The methodology is based on the use of two concurrent expert systems [ 96 , 97 , 98 ] for the calculation of the radon risk value, together with a decision tree-based regression model, that will determine the final recommendations associated to such risk. All of them are respectively labelled in Figure 1 as ‘Fuzzy inference system Fc’, ‘Fuzzy inference system RR’ and ‘Regression Tree #Number of Tree’.…”

“…The representation of antecedents and consequent is carried out by means of the well-known membership functions—the mathematical representation of a fuzzy set—that determine the degree of membership of a certain value to a particular set. The subsequent combination and aggregation of antecedents and consequents uses different specific operators that ensure the graphical combination of the membership functions [ 96 , 98 , 102 ].…”

“…After a relevant number of data collection cycles, it is possible to establish the correlation coefficient that exists between the radon concentration level and the different environmental and atmospheric variables, from the values stored into the database. After that, processing is carried out on the data collected by the two Mamdani-type fuzzy-logic expert systems that work in a concurrent way [ 96 , 97 , 98 ]. The first system is provided with a feedback from the historian as Figure 1 shows, in which depending on the correlation existing between the radon concentration level and the value of the different atmospheric variables, the values of the different rules of the system itself are weighted in one or another way.…”

“…So far the correction factor and the radon risk have been calculated, obtained after the defuzzification of their corresponding inference systems, which work concurrently [ 96 , 97 , 98 ]. Such information must be structured into a knowledge base, together with the values of the atmospheric variables, the radon concentration, and the level of each one of the corrective measures associated to each dataset (variables and concentration values).…”

Design and Definition of a New Decision Support System Aimed to the Hierarchization of Patients Candidate to Be Admitted to Intensive Care Units

“…The developments and advances in artificial intelligence algorithms have allowed the reasoning and learning models to give support to decision support systems [ 29 ], thus allowing them to diversify and complement their usage. Starting from—either symbolic or statistical—inferential processes, approaches such as machine learning, deep learning or expert systems have the inherent capability to find probabilistic and/or logical relationships that allow for limiting the uncertainty and reducing the risk associated with decision making [ 29 , 30 , 31 , 32 , 33 , 34 , 35 ]. However, the use of these techniques requires advanced learning models, together with complex representations of reasoning or learning that implies the need for the availability of a large starting dataset, or for experts in charge of processing such data and elaborate association rules.…”