Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review

Samuel Cardoso

1,2 a

, Juliano Buss

2 b

, Javier Gomez

2 c

, Helida Santos

3 d

, Giancarlo Lucca

4 e

Adenauer Yamin

2 f

and Renata Reiser

2 g

Federal Institute of Education Science and Technology Sul-rio-grandense (IFSul), Av. Paul Harris, 97574-360, Brazil

Federal University of Pelotas (UFPel), R. Gomes Carneiro, 96010-610, Brazil

Center for Computational Sciences (C3), Federal University of Rio Grande (FURG), Av. It

alia, km 8, 96203-900, Brazil

CCST, Catholic University of Pelotas (UCPel), R. Gonc¸alves Chaves, 96015-560, Brazil

Keywords:

Fuzzy Logic, Machine Learning, Neonatal EEG, Hypoxic-Ischemic Encephalopathy, Seizure Detection,

Neonatal Neurological Monitoring, Biomarkers for Neonatal Brain Injury, Systematic Review.

Abstract:

Machine learning has advanced in healthcare, aiding diagnostics, treatment, and monitoring. In neonatal

health, it helps to classify and predict conditions such as hypoxic-ischemic encephalopathy, which requires

early detection. Thus, EEG pattern analysis is key in improving the neonatal prognosis. In this work, we

present a systematic review of the literature to identify strategies currently employed to classify and predict

neonatal EEG patterns using fuzzy logic. Fuzzy logic is particularly valuable for handling uncertainties in

biological signals and improving interpretability. Five studies were selected and analyzed, focusing on apply-

ing fuzzy systems to detect epileptic events. The reviewed studies highlight techniques involving EEG data,

emphasizing the role of fuzzy logic in advancing the understanding and management of neonatal neurological

conditions, contributing to the state of the art in this critical ﬁeld.

1 INTRODUCTION

Perinatal asphyxia is one of the leading causes

of neonatal mortality, which can lead to hypoxic-

ischemic encephalopathy (HIE), a severe condition

that affects approximately 20 out of every 1,000 live

births in low- and middle-income countries (Abate

et al., 2021). HIE compromises the brain to varying

degrees, often triggering epileptic seizures within the

ﬁrst hours of life. These events not only indicate the

presence of brain injuries but also reﬂect the severity

of neurological impairment (Zhou et al., 2021). Sub-

clinical seizures are difﬁcult to detect without contin-

uous EEG monitoring and represent a signiﬁcant risk

due to the potential for cumulative neurological dam-

age. In these cases, continuous EEG monitoring is

https://orcid.org/0000-0001-5076-500X

https://orcid.org/0009-0006-3862-5104

https://orcid.org/0000-0002-8408-9748

https://orcid.org/0000-0003-2994-2862

https://orcid.org/0000-0002-3776-0260

https://orcid.org/0000-0002-7333-244X

https://orcid.org/0000-0001-9934-3115

essential for early seizure detection, allowing timely

clinical interventions that can improve the neonatal

prognosis (Glass and Shellhaas, 2019).

In the context of HIE, neonatal epilepsy is a crit-

ical neurological condition whose early detection is

challenging, as it requires constant monitoring of

brain activity signals. Epileptic events and changes

in the background patterns of electroencephalography

(EEG) signals are valuable indicators of neurological

impairment, which can help predict long-term seque-

lae or immediate life-threatening risks (Toet and Lem-

mers, 2009; Wikstr

om et al., 2012). Although con-

ventional electroencephalography facilitates the iden-

tiﬁcation of these conditions, the manual interpreta-

tion of these signals is complex and time-consuming,

requiring the continuous presence of specialists for

enhanced, real-time analysis (Wu et al., 2019).

To overcome this limitation, recent research has

focused on automating the analysis of these signals

by developing algorithms that automatically detect

epileptic seizures and classify EEG background pat-

terns in newborns (Wu et al., 2019; Montazeri et al.,

2021). These algorithms provide signiﬁcant support

for early diagnosis, enabling rapid and precise in-

840

Cardoso, S., Buss, J., Gomez, J., Santos, H., Lucca, G., Yamin, A. and Reiser, R.

Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review.

DOI: 10.5220/0013362800003929

In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 1, pages 840-847

ISBN: 978-989-758-749-8; ISSN: 2184-4992

terventions that can mitigate long-term neurological

damage (Abbasi et al., 2017). Furthermore, automa-

tion allows for the expanded use of EEG in environ-

ments where specialists are limited, such as remote

areas or smaller hospitals (Pavel et al., 2020).

Among the promising approaches for improving

these automated systems is applying fuzzy logic (Ab-

basi et al., 2014). By handling uncertainties and vari-

abilities inherent in biological signals, fuzzy logic en-

ables a more ﬂexible and robust classiﬁcation of EEG

data (G

uler and

Ubeyli, 2005; Abbasi et al., 2014;

Pavel et al., 2020). This ﬂexibility is essential, as

these signals frequently present subtle and continu-

ous variations, which are difﬁcult to categorize using

traditional binary systems (Wu et al., 2019; Montaz-

eri et al., 2021). Thus, fuzzy logic applied to neona-

tal monitoring can potentially increase sensitivity and

speciﬁcity in detecting epileptic events and classify-

ing background patterns, providing a more adaptable

and precise tool that allows specialists to better under-

stand and diagnose clinical conditions.

In this context, fuzzy logic stands out for its high

explainability in classiﬁcation systems, thanks to the

use of “IF-THEN” rules formulated in language close

to human reasoning and the intuitive representation

of concepts through linguistic variables. This ap-

proach allows clear traceability of how inputs inﬂu-

ence outputs, facilitating the interpretation of results

by specialists. In applications such as EEG analysis,

fuzzy logic enables incorporating heuristic or empiri-

cal knowledge, such as known patterns of brain activ-

ity, ensuring greater transparency and reliability in the

decision-making process (Zadeh, 1996; Ross, 2010).

On the other hand, Deep Neural Networks, while

extremely effective in complex scenarios, operate as

“black boxes”, making it difﬁcult to interpret their

decisions due to the high complexity of the models.

Despite advancements in explanation methods, these

mechanisms still lack the simplicity and clarity pro-

vided by fuzzy logic. Thus, while neural networks are

preferable for tasks requiring high performance with

large volumes of data, fuzzy logic is more suitable in

contexts where interpretability and decision reliability

are essential (Samek, 2017).

In this scenario, this paper presents a systematic

review of algorithm-based approaches for the detec-

tion of epileptic events and classiﬁcation of neona-

tal EEG background patterns, focusing on methodolo-

gies that integrate fuzzy logic to enhance the accuracy,

explainability, reliability, and adaptability of these al-

gorithms in order to improve clinical interventions.

The remainder of the paper is structured into four

sections. Section 2 outlines the deﬁnition and steps

involved in conducting a systematic review. Section

3 details the methodology employed to perform the

systematic research and discusses each stage. Section

4 presents the results, while the ﬁnal section offers the

conclusions drawn from this study.

2 SYSTEMATIC REVIEW

The systematic literature review (SLR) is a rigorous

and structured method widely used to synthesize sci-

entiﬁc evidence on a speciﬁc research question. Un-

like traditional reviews, this approach adopts a trans-

parent, replicable, and protocol-driven process, mini-

mizing biases and enhancing the reliability of conclu-

sions (Higgins et al., 2019; Liberati et al., 2009). Rec-

ognized as the gold standard in evidence-based prac-

tice, systematic reviews are extensively employed in

ﬁelds such as health, education, and social sciences,

which are essential in informing science-based prac-

tices (Kitchenham, 2004; Gough et al., 2017).

A SLR follows a rigorous methodology that en-

sures its standardization and scientiﬁc validity. To

conduct a systematic review, it is essential to estab-

lish a research protocol consisting of four main steps.

The ﬁrst step involves formulating the research ques-

tions the review aims to answer, providing a clear di-

rection for the study. In the second step, the search

strategy is deﬁned, including selecting databases and

search terms to identify and retrieve relevant articles.

The third step establishes the inclusion and exclusion

criteria for the studies, while the fourth and ﬁnal step

determines the data to be extracted, how the studies

will be characterized, and the methods for synthesiz-

ing and analyzing the data (Keele et al., 2007; Liberati

et al., 2009; Gough et al., 2017; Bramer et al., 2018;

Higgins et al., 2019).

The main advantage of a systematic review is

its well-deﬁned methodology, which greatly reduces

the risk of bias in the results. This approach en-

sures that articles are not selectively chosen to align

with the author’s personal viewpoint, promoting a

more balanced and objective synthesis of the litera-

ture (Liberati et al., 2009).

3 METHODOLOGY

The methodology employed for conducting the sys-

tematic review adhered to the approach outlined by

Keele et al. (2007) and is summarized in the ﬂowchart

shown in Figure 1. This ﬂowchart illustrates the pro-

gression of the process, during which articles iden-

tiﬁed through searches are systematically excluded

from the scope of the study. A detailed discussion of

Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review

841

Research Questions

Search Terms

Data

Base

Books

Chapters

Articles

Published

until jul, 2019

Year

Title or

Keywords

Abstract

Introduction

Conclusions

Yes

Relevant

Yes

Relevant

Yes

Relevant

Published

after

jul, 2019

Figure 1: Flowchart of the theoretical review methodology.

the criteria adopted for each of the four key stages of

the systematic review process follows in subsections

3.1, 3.2, 3.3, 3.4 and 3.5.

3.1 Research Questions

The ﬁrst step in conducting a systematic review in-

volves deﬁning the research questions that must be

answered. These questions guide the development

of search terms and keyword combinations used in

database searches. Based on the problem discussed

in the Introduction, four research questions were for-

mulated as follows:

Q1. Which predictive methods are applied to classify

background patterns in neonatal EEG signals?

Q2. Which computational approaches detect epilep-

tic and subclinical seizures in neonatal EEG sig-

nals?

Q3. Which techniques are applied to identify sleep-

wake states in neonatal EEG signals?

Q4. Which methodologies have expert systems em-

ployed to diagnose encephalopathies in new-

borns shortly after birth?

It is important to note that for each planned study,

only fuzzy logic approaches were reviewed, as the re-

search aims to assess their applicability in predict-

ing neonatal encephalopathy using biomarkers de-

rived from electroencephalography recordings.

3.2 Search Terms

To deﬁne the search terms for composing the query

string, a preliminary analysis of each term was con-

ducted using the Google Scholar database, as it is

diverse and indexes articles and works from various

sources. The study considered only the last 5 years

(2020–2024). It was progressively conducted (see Ta-

ble 1 for indexes k=1 to k=17) to identify the best

terms for the composition of the query string as the

number of articles returned increased. The combi-

nation of search terms and the number of articles re-

turned for each composition are described in Table 1.

Finally, the conjunction ‘AND’ and the term ‘fuzzy’

were added to the end of the query string that yielded

most of the results (see k=17 in Table 1) to limit the

search to articles that use prediction methods based

on fuzzy logic.

3.3 Data Bases

The Google Scholar database can be a useful tool for

exploring broad and preliminary literature or for as-

sisting in developing the query string. However, due

to its lack of quality control, limited precision, and

difﬁculty in ﬁltering results, it is not the best option as

a primary source for systematic reviews. Therefore,

specialized databases (DB) such as ACM Digital Li-

brary, EI Compendex, IEEE Xplore, Web of Science,

PubMed, ScienceDirect, Scopus, and SpringerLink

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

842

Table 1: Keyword combinations used to construct the search string and the respective number of articles returned.

K Combination

Articles

returned

1 (“baby”) AND (“electroencephalogram”) 4440

2 (“preterm”) AND (“electroencephalogram”) 4840

3 (“newborn”) AND (“electroencephalogram”) 6940

4 (“neonatal”) AND (“electroencephalogram”) 12200

5 (“preterm” OR “baby”) AND (“electroencephalogram”) 8090

6 (“preterm” OR “newborn”) AND (“electroencephalogram”) 9130

7 (“baby” OR “newborn”) AND (“electroencephalogram”) 9700

8 (“preterm” OR “neonatal”) AND (“electroencephalogram”) 13400

9 (“neonatal” OR “baby”) AND (“electroencephalogram”) 14300

10 (“neonatal” OR “newborn”) AND (“electroencephalogram”) 14400

11 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“electroencephalogram”) 16200

12 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“electroencephalography”) 16400

13 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“EEG”) 20900

14 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“EEG” OR “electroencephalogram”) 22500

15 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“EEG” OR “electroencephalography”) 23100

16 (“neonatal” OR “newborn” OR “baby” OR “preterm”) AND (“EEG” OR “electroencephalography” OR “electroencephalogram”) 26000

17 (query string presented in k=16) AND (“fuzzy”) 2730

were selected, as they collectively provide a more ro-

bust, reliable, and structured approach for conducting

systematic reviews.

3.4 Exclusion/Inclusion Criteria

Exclusion Criteria (EC) are deﬁned as factors that,

while meeting the inclusion criteria, possess addi-

tional characteristics that may hinder the study’s suc-

cess or lead to the inclusion of irrelevant or unneces-

sary information. These criteria are designed to en-

sure the study remains focused and reliable. The EC

considered in this study were:

• EC1: Books and book chapters, as the focus is

on identifying recent developments by researchers

rather than exploring the concepts and deﬁnitions

related to the subject discussed in the work;

• EC2: Articles published before July 2019, as the

goal is to focus on the most recent developments

and studies being conducted by scholars and re-

searchers;

• EC3: Articles that do not incorporate electroen-

cephalography or electrocorticography (ECoG)

signals in their methodology, as the focus of this

systematic review is on the utilization of brain-

derived signals to assess the clinical status of new-

born patients;

• EC4: Articles that did not demonstrate the use of

fuzzy logic in the methodology for analyzing EEG

signals.

Inclusion Criteria (IC) are deﬁned as the main

characteristics of the population or research being

conducted. ICs are used to answer the research ques-

tions and are presented below:

• Title and Keywords: Does the article title reﬂect

the application of expert systems for predicting or

classifying neurological disorders in newborns us-

ing electroencephalography (EEG) signal analy-

sis? Do the keywords of the article include elec-

troencephalography or terms related to this con-

text, suggesting that the article likely involves the

analysis of such signals?

• Abstract: Does the article discuss the application

of fuzzy systems for monitoring and predicting

epileptic seizures in EEG signals? Does the study

discuss the application of the fuzzy system for

classifying background patterns in EEG signals?

Does the study present a fuzzy system for classi-

fying the sleep-wake cycle in EEG signals? Does

the study present an expert system for monitoring

and diagnosis based on fuzzy logic?

• Introduction and Conclusion: Are the study’s ob-

jectives clearly deﬁned? Does it propose us-

ing fuzzy logic to predict or classify electroen-

cephalography (EEG) exams? Does the article re-

port that the designed system can be applied to the

neonatal population?

3.5 Selecting Works

After the inclusion and exclusion criteria were de-

ﬁned, the study was conducted using specialized

databases (see section 3.4) and managed with the Par-

sifal software. The ﬁrst step consisted of performing

a search based on the deﬁned research terms (see k =

17 in Table 1) across all selected databases, initially

without applying any exclusion criteria.

Subsequently, the articles were imported into Par-

sifal, where duplicate studies were identiﬁed and re-

moved. Duplicates were eliminated by retaining only

Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review

843

the articles from the database with the most records.

After the removal of duplicate studies (RDS), the

exclusion criteria were applied as follows: the publi-

cation date of the retrieved articles (EC2) and the type

of publication, speciﬁcally excluding books and book

chapters (EC1).

After applying criteria EC1 and EC2, the remain-

ing articles were evaluated based on EC3, which ex-

cluded studies that did not incorporate EEG signals in

their methodology. This evaluation was performed by

analyzing each article’s title and keywords.

Following the evaluation of the titles and key-

words of the articles, EC4 was applied by analyzing

the abstracts to determine whether the methodology

employed included fuzzy logic. Thirty-one articles

remained after applying this exclusion step.

Finally, the last step was to analyze the 31 remain-

ing articles to determine whether they addressed the

questions deﬁned by the inclusion criteria. Initially,

each article’s title and keywords were evaluated. If

the title did not meet the established criteria, the arti-

cle was discarded. However, if the title and keywords

complied with the criteria, the analysis proceeded to

the article’s abstract. If the abstract met the inclusion

criteria, the introduction and conclusion of the article

were then evaluated to conﬁrm its acceptance.

The described process is summarized in Table 2,

which presents the total number of articles retrieved

from each digital library at each study stage. Figure

2, in turn, summarizes the number of articles excluded

during the application of the inclusion and exclusion

criteria adopted in this systematic review.

6 Citations

identified

by searching the

ACM Digital

Library database

6 Citations

identified

by searching the

EI Compendex

database

9 Citations

identified

by searching the

IEEE Xplore

database

1218 Records

screened

8 Citations

identified

by searching the

PubMed

database

7 Citations

identified

by searching the

Web of Science

database

152 Citations

identified

by searching the

ScienceDirect

database

944 Citations

identified

by searching the

Scopus

database

86 Citations

identified

by searching the

SpringerLink

database

1185 Records

excluded

- 81, RDS

- 141, EC 1 and 2

- 782, EC 3

- 183, EC 4

31 Full-text

articles assessed

for eligibility

26 Full-text articles

excluded

- 10 title

- 6 abstract

- 10 introduction and

conclusion

5 Studies included

in the systematic

review

Figure 2: PRISMA Flow Diagram.

Table 2: Total articles returned for each of the described

steps.

Databases Initial

RDS

EC1

EC2

EC3 EC4 IC

ACM Digital

Library

6 0 0 0 0 0

EI Compendex 6 2 2 1 0 0

IEEE Xplorer 9 6 6 6 1 0

Web of Science 7 7 7 6 5 4

PubMed 8 0 0 0 0 0

ScienceDirect 152 134 94 22 1 0

Scopus 944 911 841 166 24 1

SpringerLink 86 77 46 13 0 0

Total 1218 1137 996 214 31 5

After applying all inclusion criteria, ﬁve articles

based on the established research questions were se-

lected (brieﬂy discussed in Sect. 4), with information

detailed in Table 3.

4 RESULTS

In this section, each of the ﬁve selected articles is

brieﬂy presented, highlighting the questions they ad-

dress and their main points.

Before starting this analysis, a word cloud was

created based on the frequency of keywords found in

each article’s titles, abstracts, introductions, and con-

clusions. This approach allowed for a more detailed

examination of the central concepts discussed in the

selected articles.

In constructing the word cloud presented in Figure

3, the most frequently appearing keywords in the se-

lected sections were considered. The most cited terms

are displayed in the ﬁgure with larger font sizes, while

less frequently cited terms across the set of articles are

shown with proportionally smaller font sizes. This

visualization facilitates the identiﬁcation of the most

relevant concepts discussed in the analyzed articles.

Figure 3: Word cloud assembled from selected articles.

Thus, it can be concluded that the most fre-

quent words precisely deﬁne the scope of this study,

namely, the terms used in the queries. Additionally,

terms such as “ECoG” and “Hypoxic-Ischemic En-

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

844

Table 3: Summary of key characteristics and ﬁndings of selected studies.

Study Methodology Results Dataset Relevance Fuzzy Logic-based Approach

Liu

et al.

(2024)

Hybrid EEG classiﬁcation using

PCANet, phase space

reconstruction (PSR), and power

spectral density (PSD). Two-

layered classiﬁcation: F-KNN for

initial layer; SVM for second

layer (Liu et al., 2024).

Accuracy: 99.4%, Sensitivity:

99.5%, Speciﬁcity: 99.75%.

Demonstrated potential for

neonatal EEG but not

validated on neonatal

signals.

Non-

neonatal

(Public)

Addressed Q2 and Q3,

highlighting seizure

prediction potential

without neonatal

validation.

Cascade Deep Learning Architecture:

Fuzzy K-Nearest Neighbor

(F-KNN)

Abbasi

et al.

(2020)

Spectral analysis using Fourier

Transform with a Type-1 fuzzy

classiﬁer (FFT-Type-1-FLC).

Focused on high-frequency

spike transients (80-120 Hz)

(Abbasi et al., 2020).

Overall performance:

98.87%. Effective in

identifying spike transients

but limited to speciﬁc

patterns (e.g., interburst

intervals not covered).

Animal

(Private)

Addressed Q1 (partially),

Q2, and Q4, providing

a strong framework for

transient detection.

Type-1 Fuzzy Logic Classiﬁers:

- FFT-Type-1-FLC classiﬁer

Abbasi

et al.

(2021)

Scalogram-based CNN (WS-

CNN) and fuzzy classiﬁers (WT-

Type-I-FLC and FFT-Type-I-FLC)

using CWT for feature

extraction. Focused on the

identiﬁcation of post-hypoxic

epileptiform EEG spikes

(Abbasi et al., 2021).

WS-CNN: Accuracy 99.81%,

AUC 1.0. Fuzzy classiﬁers

effective (99.04%, 98.42%

accuracy), but CNN

outperformed fuzzy systems

in robustness.

Animal

(Private)

Addressed Q1 (partially),

Q2, and Q4, comparing

deep learning and fuzzy

systems effectively.

Type-1 Fuzzy Logic Classiﬁer:

- Wavelet-based fuzzy classiﬁer

(WT-Type-1-FLC)

- Fast Fourier transform-based

Fuzzy classiﬁer (FFT-Type-1-FLC)

Abbasi

et al.

(2019b)

Reverse biorthogonal wavelets

(Rbio-WT-Type-1-FLC) for

detecting hypoxic-ischemic

transients in gamma range (80-

120 Hz) (Abbasi et al., 2019b).

Overall performance:

99.78%. Demonstrated

biomarkers for HIE

diagnosis during latent

phase.

Animal

(Private)

Addressed Q1 (partially),

Q2, and Q4. Signiﬁcant

potential for diagnostic

support tools.

Type-1 Fuzzy Classiﬁer:

Rbio-Wavelet Type-1 fuzzy classiﬁer

(rbio-WT-Type-1-FLC)

Abbasi

et al.

(2019a)

Wavelet-based Type-2 fuzzy

classiﬁer (WT-Type-2-FLC) for

sharp-wave transients detection

post-hypoxic-ischemic insult

(Abbasi et al., 2019a).

Identiﬁed transients

correlated with neuronal

preservation during latent

phase. Key for therapeutic

timing.

Animal

(Private)

Fully addressed Q4,

focusing on biomarkers

and neuroprotective

intervention timing.

Type-2 Fuzzy Classiﬁer:

Wavelet-based Type-2 fuzzy classiﬁer

(WT-Type-2-Fuzzy)

cephalopathy (HIE)” also stand out, even though they

were not directly included in the queries, as they are

closely related to the overall context of the study. In

this context, ECoG represents a brain-derived signal,

similar to surface EEG, while HIE refers to a clinical

condition associated with the neonatal population.

The analysis included ﬁve selected studies, each

employing fuzzy logic methodologies to enhance the

analysis of neonatal EEG signals. For instance, Liu

et al. (2020) introduced a hybrid EEG classiﬁca-

tion model that utilized PCANet, phase space recon-

struction (PSR), and power spectral density (PSD) for

seizure detection. While the model achieved high ac-

curacy, sensitivity, and speciﬁcity, it lacked valida-

tion with neonatal datasets. Similarly, Abbasi et al.

(2020) focused on high-frequency spike transients us-

ing Fourier Transform-based spectral analysis and a

Type-1 fuzzy classiﬁer (FFT-Type-1-FLC). Their ap-

proach demonstrated strong performance in detecting

speciﬁc EEG patterns but showed limited scope for

identifying other pathological background patterns.

A comparative study between fuzzy classiﬁers and

a scalogram-based CNN (WS-CNN) for identifying

high-frequency spikes is conducted by Abbasi et al.

While the fuzzy classiﬁers demonstrated robust accu-

racy, the CNN outperformed them regarding robust-

ness and adaptability to morphological variations. In

an earlier study, Abbasi et al. (2019b) proposed a

Type-1 fuzzy classiﬁer based on reverse biorthogo-

nal wavelets (Rbio-WT-Type-1-FLC), designed to de-

tect hypoxic-ischemic transients and identify valu-

able biomarkers for the early diagnosis of hypoxic-

ischemic encephalopathy. Following this same line of

research, Abbasi et al. (2019a) developed a Type-2

fuzzy classiﬁer (WT-Type-2-FLC) correlating sharp-

wave transients with neuronal survival during the la-

tent phase after hypoxic-ischemic insults, highlight-

ing its potential to guide neuroprotective interven-

tions.

Table 3 presents the key characteristics and ﬁnd-

ings of these studies, detailing their methodologies,

results, and speciﬁc contributions.

This structured analysis highlights the diverse

methodologies employed and their contributions to

the ﬁeld. While the results underscore the potential

of fuzzy logic systems, they also reveal critical gaps,

particularly the lack of validation in human neonatal

datasets and the need to address broader EEG patterns

for clinical application.

5 CONCLUSIONS

Machine learning-based methodologies have been

widely used to solve problems in the healthcare ﬁeld.

In neonatal contexts, conditions such as hypoxic-

ischemic encephalopathies and epileptic events re-

quire continuous monitoring and early diagnosis, as

they can cause severe neurological damage.

Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review

845

The objective of this study is to identify predic-

tion and classiﬁcation methodologies based on fuzzy

logic applied to the analysis of neonatal EEG signals,

focusing on advancements that improve system sen-

sitivity, speciﬁcity, and interpretability. To this end,

a systematic review was conducted to understand the

state-of-the-art techniques used for the detection of

epileptic events and background patterns in neonatal

EEGs. The review was carried out considering strict

inclusion and exclusion criteria to select relevant stud-

ies, ensuring the quality and reliability of the results.

Based on the reviewed studies, it was observed

that fuzzy logic techniques - such as Type 1 and Type

2 fuzzy classiﬁers, wavelet-based fuzzy systems, fre-

quency spectrum-based fuzzy systems, and Fuzzy K-

Nearest Neighbor (F-KNN) - are widely applied in

the detection of epileptic events and the classiﬁca-

tion of neonatal EEG background patterns. These ap-

proaches demonstrate how fuzzy logic improves the

sensitivity, speciﬁcity, and interpretability of algo-

rithms that process complex biological signals, par-

ticularly in challenging contexts such as hypoxic-

ischemic encephalopathy. Nonetheless, the valida-

tion of these techniques has predominantly been per-

formed on animal models or datasets of adult patients,

which presents limitations for their direct application

to human neonates. This methodological gap high-

lights the need for future studies aimed at validating

these techniques in real neonatal populations to en-

sure their clinical relevance.

From this perspective, several aspects remain un-

explored, offering opportunities for future research.

In this context, none of the selected studies fully ad-

dress the integration of multiple EEG patterns beyond

high-frequency peak transients. Other patterns, such

as burst suppression, excessive discontinuity, low-

voltage patterns, and inactivity, were not considered

in the reviewed studies, despite their clinical impor-

tance. Furthermore, the databases used often include

EEG signals with limited characteristics or obtained

under controlled conditions, which may not represent

the broad variability observed in neonatal intensive

care units (NICUs). Thus, developing a comprehen-

sive fuzzy system capable of identifying a wider range

of background patterns and validating by real clinical

data could signiﬁcantly improve diagnostic accuracy.

Additionally, while the reviewed studies explore

conventional fuzzy classiﬁers, none employed the

Adaptive Neuro-Fuzzy Inference System (ANFIS).

The use of ANFIS could be a promising solution as it

combines the high accuracy of neural networks with

the explainability and interpretability of fuzzy logic.

Implementing this method would enable the creation

of hybrid systems capable of automatically adjusting

fuzzy rules based on EEG patterns, thereby enhancing

both accuracy and interpretability. Nevertheless, the

lack of widely available and speciﬁc neonatal EEG

databases limits the evaluation of new methods and

underscores the need for initiatives to build standard-

ized neonatal clinical repositories. Another possibil-

ity would be the integration of real-time monitoring

systems based on fuzzy logic, providing continuous

feedback to physicians in NICUs, leading to faster

and more effective interventions.

Finally, incorporating multimodal data sources —

such as oxygen saturation and clinical observations

— into fuzzy logic-based systems could offer a more

holistic approach to neonatal neurological monitor-

ing, providing robust support for early diagnosis and

treatment planning. So, given these gaps, the next step

is to develop and implement a comprehensive fuzzy

logic-based methodology for neonatal EEG analy-

sis that addresses these unexplored areas. This ap-

proach should prioritize validation using real neonatal

datasets to ensure greater clinical applicability while

aiming to enhance the accuracy, explainability, and

adaptability of diagnostic tools, thereby offering more

reliable support for clinical interventions and improv-

ing outcomes in neonatal care.

ACKNOWLEDGEMENTS

The authors would like to thank the follow-

ing Brazilian funding agencies: CAPES, CNPq

(309160/2019-7; 311429/2020-3, 150160/2023-

2), PqG/FAPERGS (21/2551-0002057-1), and

FAPERGS/CNPq (23/2551-0000126-8).

REFERENCES

Abate, B. B., Bimerew, M., Gebremichael, B., Menge-

sha Kassie, A., Kassaw, M., Gebremeskel, T., and

Bayih, W. A. (2021). Effects of therapeutic hy-

pothermia on death among asphyxiated neonates with

hypoxic-ischemic encephalopathy: a systematic re-

view and meta-analysis of randomized control trials.

PloS one, 16(2):e0247229.

Abbasi, H., Bennet, L., Gunn, A. J., and Unsworth, C. P.

(2017). Robust wavelet stabilized ‘footprints of

uncertainty’for fuzzy system classiﬁers to automat-

ically detect sharp waves in the eeg after hypoxia

ischemia. International journal of neural systems,

27(03):1650051.

Abbasi, H., Bennet, L., Gunn, A. J., and Unsworth, C. P.

(2019a). Automatically identiﬁed micro-scale sharp-

wave transients in the early-latent phase of hypoxic-

ischemic eeg from preterm fetal sheep reveal timing

relationship to subcortical neuronal survival. In 2019

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

846

41st Annual International Conference of the IEEE En-

gineering in Medicine and Biology Society (EMBC),

pages 7084–7087. IEEE.

Abbasi, H., Bennet, L., Gunn, A. J., and Unsworth,

C. P. (2019b). Latent phase detection of hypoxic-

ischemic spike transients in the eeg of preterm fetal

sheep using reverse biorthogonal wavelets & fuzzy

classiﬁer. International Journal of Neural Systems,

29(10):1950013.

Abbasi, H., Gunn, A. J., Bennet, L., and Unsworth, C. P.

(2020). Latent phase identiﬁcation of high-frequency

micro-scale gamma spike transients in the hypoxic is-

chemic eeg of preterm fetal sheep using spectral anal-

ysis and fuzzy classiﬁers. Sensors, 20(5):1424.

Abbasi, H., Gunn, A. J., Unsworth, C. P., and Bennet,

L. (2021). Advanced deep learning spectroscopy of

scalogram infused cnn classiﬁers for robust identiﬁ-

cation of post-hypoxic epileptiform eeg spikes. Ad-

vanced Intelligent Systems, 3(2):2000198.

Abbasi, H., Unsworth, C. P., Gunn, A. J., and Bennet, L.

(2014). Superiority of high frequency hypoxic is-

chemic eeg signals of fetal sheep for sharp wave de-

tection using wavelet-type 2 fuzzy classiﬁers. In 2014

36th Annual International Conference of the IEEE

Engineering in Medicine and Biology Society, pages

1893–1896. IEEE.

Bramer, W. M., De Jonge, G. B., Rethlefsen, M. L., Mast,

F., and Kleijnen, J. (2018). A systematic approach

to searching: an efﬁcient and complete method to de-

velop literature searches. Journal of the Medical Li-

brary Association: JMLA, 106(4):531.

Glass, H. C. and Shellhaas, R. A. (2019). Acute symp-

tomatic seizures in neonates. In Seminars in Pediatric

Neurology, volume 32, page 100768. Elsevier.

Gough, D., Thomas, J., and Oliver, S. (2017). An introduc-

tion to systematic reviews.

uler, I. and

Ubeyli, E. D. (2005). Adaptive neuro-fuzzy in-

ference system for classiﬁcation of eeg signals using

wavelet coefﬁcients. Journal of neuroscience meth-

ods, 148(2):113–121.

Higgins, J., Chandler, J., Cumpston, M., Li, T., Page, M.,

and Welch, V. (2019). Cochrane handbook for system-

atic reviews of interventions. Chichester(UK): John

Wiley & Sons, 2nd edition.

Keele, S. et al. (2007). Guidelines for performing system-

atic literature reviews in software engineering. Tech-

nical report, Technical report, ver. 2.3 ebse technical

report. ebse.

Kitchenham, B. (2004). Procedures for performing sys-

tematic reviews. Keele, UK, Keele University,

33(2004):1–26.

Liberati, A., Altman, D. G., Tetzlaff, J., Mulrow, C.,

Gøtzsche, P. C., Ioannidis, J. P., Clarke, M., Dev-

ereaux, P. J., Kleijnen, J., and Moher, D. (2009). The

prisma statement for reporting systematic reviews and

meta-analyses of studies that evaluate health care in-

terventions: explanation and elaboration. Annals of

internal medicine, 151(4):W–65.

Liu, C., Chen, W., and Li, M. (2024). A hybrid eeg clas-

siﬁcation model using layered cascade deep learning

architecture. Medical & Biological Engineering &

Computing, pages 1–17.

Montazeri, S., Pinchefsky, E., Tse, I., Marchi, V., Koho-

nen, J., Kauppila, M., Airaksinen, M., Tapani, K.,

Nevalainen, P., Hahn, C., et al. (2021). Building

an open source classiﬁer for the neonatal eeg back-

ground: a systematic feature-based approach from ex-

pert scoring to clinical visualization. Frontiers in hu-

man neuroscience, 15:675154.

Pavel, A. M., Rennie, J. M., de Vries, L. S., Blennow,

M., Foran, A., Shah, D. K., Pressler, R. M., Kapel-

lou, O., Dempsey, E. M., Mathieson, S. R., et al.

(2020). A machine-learning algorithm for neonatal

seizure recognition: a multicentre, randomised, con-

trolled trial. The Lancet Child & Adolescent Health,

4(10):740–749.

Ross, T. J. (2010). Fuzzy Logic with Engineering Applica-

tions. John Wiley & Sons, Ltd.

Samek, W. (2017). Explainable artiﬁcial intelligence: Un-

derstanding, visualizing and interpreting deep learn-

ing models. arXiv preprint arXiv:1708.08296.

Toet, M. C. and Lemmers, P. M. (2009). Brain monitoring

in neonates. Early human development, 85(2):77–84.

Wikstr

om, S., Pupp, I. H., Ros

en, I., Norman, E., Fell-

man, V., Ley, D., and Hellstr

om-Westas, L. (2012).

Early single-channel aeeg/eeg predicts outcome in

very preterm infants. Acta paediatrica, 101(7):719–

726.

Wu, D., Wang, Z., Jiang, L., Dong, F., Wu, X., Wang, S.,

and Ding, Y. (2019). Automatic epileptic seizures

joint detection algorithm based on improved multi-

domain feature of ceeg and spike feature of aeeg.

IEEE Access, 7:41551–41564.

Zadeh, L. A. (1996). Fuzzy logic= computing with words.

IEEE transactions on fuzzy systems, 4(2):103–111.

Zhou, K. Q., McDouall, A., Drury, P. P., Lear, C. A., Cho,

K. H., Bennet, L., Gunn, A. J., and Davidson, J. O.

(2021). Treating seizures after hypoxic-ischemic en-

cephalopathy—current controversies and future direc-

tions. International Journal of Molecular Sciences,

22(13):7121.

Fuzzy Logic for Neonatal EEG Analysis: A Systematic Review

847