Improvement of newborn screening using a fuzzy inference system
2017
Newborn screening plans are a routine for newborn care.We present a fuzzy inference system to support newborn screening.It screens for 46 inborn errors of metabolism and takes as input up to 42 analytes.It has been tested with two different types of samples from three distinct sources.High accuracy on diagnosis is obtained in 98.7% less time than laboratory personnel. This paper presents a decision support system (DSS) called DSScreening to rapidly detect inborn errors of metabolism (IEMs) in newborn screening (NS). The system has been created using the Aide-DS framework, which uses techniques imported from model-driven software engineering (MDSE) and soft computing, and it is available through eGuider, a web portal for the enactment of computerised clinical practice guidelines and protocols.MDSE provides the context and techniques to build new software artefacts based on models which conform to a specific metamodel. It also offers separation of concern, to disassociate medical from technological knowledge, thus allowing changes in one domain without affecting the other. The changes might include, for instance, the addition of new disorders to the DSS or new measures to the computation related to a disorder. Artificial intelligence and soft computing provide fuzzy logic to manage uncertainty and ambiguous situations. Fuzzy logic is embedded in an inference system to build a fuzzy inference system (FIS); specifically, a single-input rule modules connected zero-order Takagi-Sugeno FIS. The automatic creation of FISs is performed by the Aide-DS framework, which is capable of embedding the generated FISs in computerized clinical guidelines. It can also create a desktop application to execute the FIS. Technologically, it supports the addition of new target languages for the desktop applications and the inclusion of new ways of acquiring data.DSScreening has been tested by comparing its predictions with the results of 152 real analyses from two groups: (1) NS samples and (2) clinical samples belonging to individuals of all ages with symptoms that do not necessarily correspond to an IEM. The system has reduced the time needed by 98.7% when compared to the interpretation time spent by laboratory professionals. Besides, it has correctly classified 100% of the NS samples and obtained an accuracy of 70% for samples belonging to individuals with clinical symptoms.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
58
References
8
Citations
NaN
KQI