Shahmoradi et al.

BMC Women's Health (2024) 24:234 BMC Women's Health

https://doi.org/10.1186/s12905-024-03074-3

RESEARCH Open Access

Nutritional management recommendation

systems in polycystic ovary syndrome:
a systematic review
Leila Shahmoradi1, Leila Azadbakht2,3, Jebraeil Farzi4, Sharareh Rostam Niakan Kalhori1, Alireza Banaye Yazdipour1,5,6*
and Fahimeh Solat1,7*

Abstract
Background People with polycystic ovary syndrome suffer from many symptoms and are at risk of developing
diseases such as hypertension and diabetes in the future. Therefore, the importance of self-care doubles. It is mainly to
modify the lifestyle, especially following the principles of healthy eating. The purpose of this study is to review artificial
intelligence-based systems for providing management recommendations, especially food recommendations.
Materials and methods This study started by searching three databases: PubMed, Scopus, and Web of Science,
from inception until 6 June 2023. The result was the retrieval of 15,064 articles. First, we removed duplicate studies.
After the title and abstract screening, 119 articles remained. Finally, after reviewing the full text of the articles and
considering the inclusion and exclusion criteria, 20 studies were selected for the study. To assess the quality of articles,
we used criteria proposed by Malhotra, Wen, and Kitchenham. Out of the total number of included studies, seventeen
studies were high quality, while three studies were moderate quality.
Results Most studies were conducted in India in 2021. Out of all the studies, diagnostic recommendation systems
were the most frequently researched, accounting for 86% of the total. Precision, sensitivity, specificity, and accuracy
were more common than other performance metrics. The most significant challenge or limitation encountered in
these studies was the small sample size.
Conclusion Recommender systems based on artificial intelligence can help in fields such as prediction, diagnosis,
and management of polycystic ovary syndrome. Therefore, since there are no nutritional recommendation systems
for these patients in Iran, this study can serve as a starting point for such research.
Keywords Polycystic ovary syndrome, Artificial intelligence, Application, Decision support system, Nutrition
recommender system

*Correspondence:
Alireza Banaye Yazdipour
[email protected]; [email protected]
Fahimeh Solat
[email protected]
Full list of author information is available at the end of the article

© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included
in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The
Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available
in this article, unless otherwise stated in a credit line to the data.
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 2 of 26

Introduction images, segmentation, and classification methods.

Polycystic ovary syndrome affects 8 to 13% of women of The evaluation highlighted that the Bayesian classi-
reproductive age worldwide, making it the most common fier achieved the highest accuracy of 93.93%. This study
endocrine problem. This condition can cause menstrual underscores the significant potential of AI in diagnosing
disorders, lack of ovulation, obesity, acne, hirsutism, hair PCOS and recommends further research to implement
loss, and baldness. Long-term complications include this technology effectively [11].
endometrial cancer, infertility, insulin resistance, type 2 Boyle et al. conducted a study in Australia in 2018 to
diabetes, high blood pressure, heart disease, depression, assess the need for assistance among individuals with
and stress. Short-term consequences and complications PCOS and evaluate mobile health applications in this
may also arise [1]. Early diagnosis of the disease is very area. The results show that 98% of participants owned
important and can reduce the duration of the disease and smartphones, 72% had previously used an application for
the mortality rate. Studies show that in many diseases, self-care, and 91% expressed willingness to use a PCOS-
early diagnosis is difficult for health care providers. On specific app if available for managing this syndrome.
the other hand, patients do not have the information Accurate, evidence-based information was deemed
related to self-management and do not have the necessary essential in this study, and all the assessed applications
knowledge to obtain this information. Social withdrawal met the required quality standards [12].
may become more prevalent as physical symptoms like The research conducted by Portugal et al. in Canada in
acne, hair loss, and depression manifest [2–4]. Follow- 2017 focused on the use of machine learning techniques
ing diagnosis, implementing lifestyle modifications to in recommender systems. The study aimed to identify
address symptoms like elevated cholesterol and insu- associated issues and assist researchers in implementing
lin resistance is recognized as an innovative therapeutic these systems more effectively. The findings of the study
approach. Given the prevalent overweight status among highlighted various machine learning techniques and
women with this syndrome, the significance of adhering their applications, as well as primary and alternative per-
to a nutritious diet and engaging in physical activity has formance criteria [13].
been underscored. Consequently, the imperative of devis- In 2019, Abhari et al. conducted a study in Iran to
ing strategies to promote adherence to healthy dietary assess the characteristics of nutritional recommender
habits and facilitate weight loss among affected individu- systems. The study revealed that if these systems are
als is deemed essential and inevitable [1]. properly designed, implemented, and evaluated, they can
The use of information technology greatly facilitates serve as effective tools to improve nutrition and promote
the prevention, diagnosis and treatment of chronic dis- a healthy lifestyle [14].
eases and increases their accuracy. Among these tech- The study conducted in India in 2021 by Kaur and col-
nologies, it can be mentioned decision support systems, leagues aimed to develop a method for classifying food
mobile-based applications, virtual reality, augmented images to track patients’ meals and provide guidance to
reality, and intelligent decision-making systems. These nutritionists on recommended tactics and image classi-
technologies largely solve the challenge of accessing data fication. The research focused on utilizing deep learning
and evidence-based information for both patients and approaches, particularly convolutional neural networks
medical providers [5–10]. (CNNs), to classify Indian food images accurately [15].
In the healthcare domain, recommender systems rep- After conducting research, we found no systematic
resent a practical technology enabling self-care through review that assesses nutritional recommendations for
tailored recommendations. Ultrasound imaging can sup- individuals with PCOS. Given the significance of utiliz-
port early disease detection, alleviating healthcare pro- ing companion health and AI in managing chronic condi-
vider workload and expediting diagnosis. This strategy tions like PCOS, this study aims to explore the effects and
not only conserves resources and reduces expenses but uses of AI-based systems for PCOS.
also leverages mobile phone platforms to enhance aware-
ness, disseminate evidence-based knowledge, and foster Motivation
beneficial shifts in individual behaviors and habits. By The advancement of AI in healthcare has made it nec-
harnessing artificial intelligence, a recommender sys- essary to use AI-powered recommender systems for
tem can enhance dietary practices among patients with predicting, diagnosing, treating, and managing chronic
PCOS, leveraging the widespread adoption of smart- ailments like PCOS. There are several reasons why this
phones to improve access to medical professionals and is important. Ultrasound images are necessary to diag-
reliable information [2, 4]. nose PCOS. However, inaccuracies in counting follicles,
The study conducted by Jan et al. in India in 2022 ana- high diagnostic test costs in developing countries, time-
lyzed six AI models for diagnosing PCOS. They com- consuming tests, doctor workload, and diagnostic errors
pared these models based on the number of ultrasound can lead to inaccurate diagnoses. AI techniques can
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 3 of 26

automatically diagnose diseases using ultrasound images, Inclusion criteria:

overcoming challenges [16–18]. Additionally, since this
disease is intricate and has no definitive treatment, the 1) Studies about recommender systems
current approach involves a combination of medication 2) Articles related to PCOS
and lifestyle changes for disease management. Hence, 3) Articled related to diet management in PCOS
utilizing recommender systems or self-care systems that 4) Studies that developed an AI system or application
prioritize healthy nutrition could be beneficial in mitigat-
ing symptoms and lessening the likelihood of associated Exclusion criteria:
mental health issues [19, 20]. Besides disease prediction
and probability estimation, recommender systems can 1) Review articles, meta-analysis, conference abstracts,
detect suspicious cases based on AI and take action to letters to the editor, book chapter.
prevent disease occurrence or early detection [21]. 2) Articles that are unrelated to the goals of the current
research.
Contribution 3) Articles whose full text is written in non-English
The article aims to explore the use of artificial rec- language.
ommender systems in polycystic ovary syndrome 4) Articles whose full text is not available for data
research. We also examine the challenges and limita- extraction.
tions of using these systems and algorithms. This article
attracts researchers to conduct studies in the field of
recommender systems. The sections of this article are as Study selection
follows: In this stage, one of the researchers entered all the arti-
cles retrieved from the three databases (PubMed, Web
•  Review of studies from the perspective of of Science, and Scopus) into the Endnote X9 (Thomson
publication. Reuters, Toronto, Ontario, Canada) software. Another
•  Review studies of characteristics. researcher removed duplicates. Two of researchers sepa-
•  Examining the limitations and challenges of rately, checked the title and abstract of the articles. The
designing recommender systems. items that did not match the inclusion and exclusion
criteria were excluded from the study. Finally, by study-
ing and examining the full text of the remaining articles,
Materials and methods he selected the articles related to the topic as the final
Study design articles. In cases where the two researchers had differing
The current systematic study was designed and imple- opinions, the supervisor provided the final decision.
mented based on the Preferred Reporting Items for
Systematic Reviews and Meta-Analyses (PRISMA) Data extraction
statement. A researcher created an Excel form with the guidance
of our supervisor. The research team reviewed articles
Data sources and extracted necessary data elements. The Excel form
In this study, we were six researchers. The first one deter- included data such as the number of articles, publication
mined the search strategy and performed the search in year and country.
PubMed, Scopus, and Web of Science databases from
inception until 6 June 2023. To conduct a search for rele- Quality assessment
vant articles, we utilized a combination of keywords from Two independent reviewers assessed the quality of stud-
the article abstracts, including “polycystic ovary syn- ies using the Newcastle-Ottawa quality assessment cri-
drome”, “recommender system”, “application program”, teria proposed by Malhotra [17], Wen et al. [16], and
“artificial intelligence” and “nutritional program”. We also Kitchenham et al. [18]. The quality assessment criteria
incorporated Medical Subject Headings (Mesh) into our consist of eleven questions: Q1) Are the aims of the study
search strategy. For the complete search strategy, please clearly defined?; Q2) Are all study questions answered?;
refer to the supplementary file provided (Table S1-S3). Q3) Are the variables used in the study clearly stated?;
Q4) Are AI techniques, such as machine learning, clearly
Selection criteria defined?; Q5) Is the data set size appropriate?; Q6) Is the
One of the researchers determined the inclusion and data collection method clearly stated?; Q7) Is the study
exclusion criteria based on similar studies, and finally methodology repeatable?; Q8) Are the results and find-
criteria were approved by the supervisor’s opinion. Inclu- ings clearly presented?; Q9) Are the performance mea-
sion and exclusion criteria are as follows: sures used to assess the model(s) clearly stated?; Q10)
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 4 of 26

Are the limitations of the study stated?; Q11) Does the Science, and Scopus databases. After removing duplicates
research have value for the academic or industry commu- (n = 6537), We took three steps: (1) screening the article
nity? The questions were ranked based on three values: titles, (2) reviewing the article abstracts, and (3) review-
“Yes = 2”, “Partial = 1”, or “No = 0”. Each study could obtain ing the full text of the articles and extracting the data by
a maximum score of 22 and a minimum score of 0. Cri- the second group. Based on the predetermined criteria,
teria used to rank the quality assessment of each study we eliminated 8,408 studies during the one and two-stage
include: i) ≤ 49% = Low quality; ii) 50% and 69% = Moder- process. In the third stage, from 119 articles unrelated
ate quality; iii) above 70% = High quality. studies (n = 86), articles with unavailable full text (n = 8),
review studies (n = 4), and book chapters (n = 1) were
Results excluded. Finally, 20 articles entered the third stage, i.e.,
Study selection a review of the full text of the articles. The research team
Figure 1 shows the process of searching and selecting extracted required data elements such as publication
articles based on the PRISMA flowchart. We found a year, country, journal/conference, purpose, study design,
total of 15,064 articles by searching in PubMed, Web of sample size, sample age range, results, tools, challenges/

Fig. 1 PRISMA flow diagram indicating results of identification and screening process for included and excluded papers
Table 1 Summary of included contributions (n=20)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges andstudy algorithms target
tion limitationDiet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Lehtin- 1997 Finland * Case Com- Pa- 27 + TPFFN Small * SOM, *
en et al. study paring tients: 33 + accuracy sample data TPFFN,
[21] the 54 was bet- volume MLP
perfor- Control ter than
mance group: SOM.
Shahmoradi et al. BMC Women's Health

of SOM 29
and
TPFFN
in an-
ticipat-
ing the
possi-
(2024) 24:234

bility of
PCOS
Zhang 2010 USA * RCT Con- 418 Clomiphene ci- Clomi- * Decision *
et al. struc- trate: 27.9 + 4.0 phene trees
[22] tion of The combina- citrate
classifi- tion of clomi- alone is
cation phene citrate better
models and metformin: and
for the 28.3 + 4.0 superior
antici- to the
pation other
of the two
occur- meth-
rence of ods for
ovula- treating
tion PCOS.
treat-
ment in
women
with
PCOS
Page 5 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Mehro- 2012 India * Original de- Normal: Normal: Bayesian Need to * Bayesian *
tra et al. scribed 150 32.24 ± 2.02 classi- improve Classifier,
[2] a abnor- Abnormal: fier gives accuracy by Multivari-
method mal: 50 31.24 ± 2.48 higher using other ate LR
that accuracy classifiers
Shahmoradi et al. BMC Women's Health

Enables than
auto- logistic
matic regres-
diagno- sion.
sis of Using
PCOS the
(2024) 24:234

based proba-
on bilistic
features model
helps
doc-
tors to
screen
early
patients
who are
more
likely to
develop
the
disease.
Page 6 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Rethina- 2016 India * Original Propos- 31 SQL The * NFRS, ANN *
valli et ing a MATLAB R 2016 a structure
al. [23] new Dataset: Polycystic based
combi- Ovarian Syndrome on fuzzy
natorial Proliferative Phase logic can
Shahmoradi et al. BMC Women's Health

struc- Endometrial Cell be used

ture to Types in risk
discov- anticipa-
er the tion
severity The
of the severity
(2024) 24:234

disease of the
in disease
people was im-
with proved.
the The pro-
disease posed
model
per-
formed
better
than the
other
created
models
with an
accu-
racy of
93.64%
Page 7 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Cahyo- 2017 Indonesia * Original Design- Patient: 3D matrix Auto- * CNN *
no1 et ing and 40 Softmax matic
al. [24] creat- Healthy: Loss function classifi-
ing a 14 Dropout cation of
system SGD method images
Shahmoradi et al. BMC Women's Health

based F1-Measure into two

on Micro-average catego-
convo- F1-Measure ries, sick
lutional and
neural healthy,
network by the
(2024) 24:234

to de-
classify signed
ultra- system
sound It was
images done
into well and
two cat- was very
egories, accurate
sick and
healthy
Page 8 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Dewi et 2018 Indonesia * Original System Gabor Wavelet The use * Competi- *
al. [25] design method of com- tive Neural
based petitive Network
on ma- neural
chine network
Shahmoradi et al. BMC Women's Health

learning can
and AI increase
to help the
Doc- accu-
tors can racy of
diag- diagno-
(2024) 24:234

nose sis in this

the article
disease The
more highest
easily accuracy
through is esti-
ultra- mated at
sound 80.84%.
images Accord-
ing to
the re-
sults, the
number
of
adopted
features
has a
direct
relation-
ship
with
accuracy
Page 9 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Thu- 2018 Indonesia * Original System 16–32 Gabor Wavelet The best More data * Elman *
failah et design features method accuracy for training Neural
al. [26] based of using affects the Network,
on the the el- time of Polynomial
Gibber- emental diagnosis SVM,
Shahmoradi et al. BMC Women's Health

Violet neural Radial Basis

method network Function
to was SVM,
extract 78.1%, Linear SVM
features which
and was
(2024) 24:234

Helping achieved
to di- with 32
agnose features.
and A higher
classify number
disease of data
for train-
ing the
network
can
increase
the
accuracy
of the
network
Page 10 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Vikas et 2018 India * Original Identify 119 18–22 PCOS Dataset Using The data set * * Apriori *
al. [27] recur- source: the used is not algorithm
ring https:// men- enough.
pat- github.com/ tioned In addition,
terns PCOS-Survey/ algo- Patients’
Shahmoradi et al. BMC Women's Health

among PCOSData rithm to concerns

the Frequent Itemset extract about
symp- Mining (FIM) the main information
toms of Spss widgets disclosure
PCOS Here,
patients the main
(2024) 24:234

using signs
a set of have
fre- per-
quently formed
used well for
items anticipa-
tion as
well as
deter-
mining
relation-
ships
between
features
Page 11 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Denny 2019 India * Original Design- patients: 18–40 SPSS V 22.0 Among * NB, LR, * *
et al. ing and 177 Principal Com- the algo- KNN, CART,
[28] creat- Healthy: ponent Analysis rithms RF, SVM
ing a 364 (PCA) used,
system Spyder Python Algo-
Shahmoradi et al. BMC Women's Health

based IDE rithm

on AI HTML with SQL RF per-
for as- for designing a formed
sistance proper user best
To di- interface with
agnose 89%
(2024) 24:234

and an- accu-

ticipate racy. The
PCOS system
disease de-
signed
accord-
ing to
experts
can be
useful
in early
disease
diagno-
sis and
save
time.
Page 12 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Thakre 2020 India * Original Design 30 Jupyter Notebook This * * RF, LR * *
et al. and features Python system Linear SVM,
[18] build helps in Radial SVM,
system the early KNN,
based diagno- Gauss-
Shahmoradi et al. BMC Women's Health

On AI sis and ian Naive

for help predic- Bayes
to di- tion of
agnose PCOS,
and an- and the
ticipate RF algo-
(2024) 24:234

PCOS rithm is
disease the most
accurate
and
reliable
algo-
rithm
with an
accuracy
of 90.9.
Abu 2021 Lebanon * Original Design- Patients: 18–40 “Polycystic Ovary The best Despite high * SFFS, LR, *
Adla et ing a 177 Syndrome”dataset, perfor- accuracy in DT, NB,
al. [29] pro- Healthy: ML application mance automatic Linear SVM,
posed 364 was model Polynomial
model related recognition SVM,
for au- to the Sugges- Radial Basis
tomatic linear tions did Function
diagno- support not show SVM,
sis of vector good perfor- Linear Dis-
PCOS ma- mance in criminant
chine, recall Classifier,
which Quadratic
was 90% Discrimi-
accurate nant, RF
with 24.
Page 13 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Hassan 2020 India * Original Design 42 R-language Among * LR, SVM *
et al. and variables R libraries: the 5 al- NB, CART,
[30] build e1071, CARET, gorithms RF
system naivebayes, rpart, used, RF
based randomForest, algo-
Shahmoradi et al. BMC Women's Health

on AI klaR, ggplot2 rithm

for help and
to di- support
agnose vector
PCOS machine
and respec-
(2024) 24:234

com- tively
pare Accu-
the racy of
perfor- 96% and
mance 95% per-
of dif- formed
ferent better.
algo-
rithms
Kodi- 2021 India * Original Design- 624 Patients under Questionnaire, The use * D-Tree, *
palli et ing a 25 K10 tool, of the KNN,
al. [31] model matplotlib, system is SVM, Fuzzy
for Fuzzy TOPSIS cost-
disease effective.
antici- The
pation perfor-
and mance
related of SVM
mental and
disor- fuzzy al-
ders gorithms
based was
94.01%
and
98.2%,
respec-
tively.
Page 14 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Song et 2022 China * Original This 721 U-Net network, A non- Ambigui- * CNNs: V3, *
al. [32] study convolutional invasive ties in the Vgg16, and
pro- block attention method, images, Vgg19
posed a module (CBAM), The There is a
model multi-instance accuracy need to con-
Shahmoradi et al. BMC Women's Health

based (MIL), MLP, of this duct more

on Resnet18 method studies to
Artificial was esti- generalize
intel- mated at the results
ligence 0.978%.
algo-
(2024) 24:234

rithm,
which is
a non-
invasive
method
with
the
help of
cap-
tured
images
It was
from
the
eyes to
help di-
agnose
PCOS.
Page 15 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Mandal 2021 India * Original Provid- 19 histogram This To deter- * K-means *
et al. ing an equalization method mine the clustering
[16] auto- can exact shape
mated auto- and size of
diag- matically the follicles
Shahmoradi et al. BMC Women's Health

nostic detect There

ap- the fol- are more
proach licles features that
for Ultra- need to be
Detec- sound considered.
tion of images
(2024) 24:234

follicles are ef-

in the fective in
ovary reduc-
using ing the
ultra- work-
sound load of
(US) doctors.
images
during
infer-
tility
treat-
ment.
Page 16 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Nilofer 2021 India * Original Present- Wiener filter, The pro- Further * Fuzzy *
et al. ing a Takagi–Sugeno– posed research is logicis,
[33] pro- Kang (TSK), com- needed Hybrid,
posed fuzzy inference bined to be done Intelligent
method method, model by institu- Water Drop
Shahmoradi et al. BMC Women's Health

for Maximum Likeli- had 99% tions and (IWD),

auto- hood (ML), accuracy stakeholders KNN,
matic Extreme Learning in de- to confirm SVM
division Adaptive Neuro- tecting the model.
of areas inference System follicles.
in ultra- (ELANFIS)
(2024) 24:234

sound
images
into
areas
with
follicles
and
without
follicles.
Page 17 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Zhang 2021 China * Original Design- Thou- DisGeNET, The * CNN, GCN *
et al. ing a sans of GWAS Catalog, current
[34] system genetic GTEx Portal algo-
based variants rithm
on in the
Shahmoradi et al. BMC Women's Health

deep field of
learning predict-
for the ing the
antici- relation-
pation ship of
of dis- disease
(2024) 24:234

eases with
related genetics
to ge- com-
netics pared
includ- with
ing algo-
PCOS rithms
Classics
such as
RF and
Support
Vector
Machine
per-
formed
better.
Page 18 of 26
Table 1 (continued)
Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
Hosain 2022 Bangla- * Obser- Devel- Dataset Image Data The * CNN, *
et al. desh vational opment 1: 1730 Generator, present InceptionV3
[35] study of a images Keras system
system Dataset not only
called 2: 339 per-
Shahmoradi et al. BMC Women's Health

PCONet images formed

To help well in
diag- diagnos-
nose ing the
pcos disease
through through
(2024) 24:234

convo- images,
lutional but also
neural per-
network- formed
based better
ultra- with an
sound accuracy
images of 98.12.
Zigarelli 2022 United * Retro- devel- 541 20–48 Rotterdam criteria The pre- The sample * K-Means *
et al. States of spective oping PCA Method diction was drawn Clustering,
[36] America study self- accuracy from a CatBoost
diag- was es- specific model
nostic timated population
predic- to be in India
tionmod- 87.5 to from several
els for 90.1% hospitals.
PCOS in
po-
tential
patients
and
clinical
provid-
ers
Page 19 of 26
Table 1 (continued)
Shahmoradi et al. BMC Women's Health

Author Pub- Country Journal conference study Study Sample Sample Tool Results Chal- Relevance to the AI (AI) System
(Ref.) lica- design aim(s) size description lenges and study algorithms target
tion limitation Diet AI Application an- di-
year tici- ag-
pa- nos-
tion tic
(2024) 24:234

Nsugbe 2023 England * Original Design- Patients: Kaggle website SVM per- More * DT, LDA, *
et al. ing and 177 formed samples LR, KNN,
[37] creat- Healthy: better with more SVM
ing a 364 than diverse data
decision other for
support used presenting
system algo- the model in
based rithms. the clinical
on AI to environ-
diag- ment is
nose needed
PCOS
and de-
termine
the
stage
of the
disease
Page 20 of 26
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 21 of 26

Fig. 2 Distribution of studies by country and year

Fig. 3 Distribution of studies by country

limitations, Relevance to the study and system target. We Distribution of articles based on journal/conference
recorded the extracted data in an Excel (Table 1). name, publisher and impact factor The articles were
appeared in 11 different journals and seven conferences.
Publication analysis The journal “Frontiers in Endocrinology” had the highest
Distribution of studies by year The studies were con- number (n = 2), while all the other journals had only pub-
ducted from 1997 to 2023. Figure 2 shows the results. 2021 lished 1 article. All conferences, except the “International
(n = 5) has the most frequency. The second frequency was Conference on Data and Information Science,” presented
related to 2018 and 2022. (n = 3). a paper on the topic. Tables 2 and 3 display the distribu-
tion of articles in this field.
Distribution of studies by country The reviewed stud-
ies were conducted in eight different countries. The fre- Study specifications
quency of these studies in each country is depicted in Frequency of studies based on AI/application Based
Figs. 2 and 3. India had the highest number of studies on the survey, most (95%) of the studies focused on mod-
(n = 9), followed by Indonesia (n = 3), the United States of els and systems utilizing AI technology [2, 16, 21–37],
America, and China (each with 2), ranking second and while only one study resulted in the creation of an AI-
third, respectively. based application [18].
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 22 of 26

Table 2 Distribution of articles based on journal/conference name, publisher and impact factor
Journal/conference name Cite Publisher Indexed in IF Count
Score quartile (ISI, Scopus, PubMed) of
papers
Human Reproduction Q1 ProQuest ISI, Scopus, PubMed 6.1 1
International Journal of Circuit Theory and Applications (I J C T A) Q1 Wiley ISI, Scopus 2.3 1
Information Systems Design and Intelligent Applications - - - - 1
Bioscience Biotechnology Research - - - - 1
International Journal of Computer Applications Q2 Other ISI, Scopus 1.1 1
Webology Other 1
Frontiers in Cell and Developmental Biology Q2 Other ISI, Scopus, PubMed 5.5 1
Journal of Medical Internet Research (JMIR) Q1 Other ISI, Scopus, PubMed 7.4 1
Healthcare Analytics - Other Scopus 1

Table 3 Distribution of articles based on conference name

Conference name Count of Frequency of studies based on the type of system
papers application: When it comes to systems and applications
IEEE India Conference (INDICON) 1 designed for various purposes, they can be classified into
International Conference on Information and Communica- 1 three types: prediction, diagnosis, and management.
tion Technology (ICOICT) Among all the conducted studies, 60% focused on diag-
International Conference on Data and Information Science 2 nosis [2, 16, 24–26, 29, 30, 32, 33, 35–37], 30% on pre-
IEEE Region 10 International Conference Tencon 1 diction [21–23, 27, 31, 34], and none on management.
International Conference on Advances in Biomedical Engi- 1 Additionally, 10% of studies focused on both diagnosis
neering (ICABME)
and prediction [18, 28]. However, none of the studies
Proceedings Of International Conference on Frontiers in 1
examined the role of nutrition in managing PCOS.
Computing and Systems
International Conference on Engineering and Emerging 1
Technologies (ICEET) Specifications of performance metrics for model eval-
uation Table 4 displays the metrics utilized in the arti-
Table 4 Performance metrics for model evaluation
cles. MAE [23], RMSE [23, 33], and RRSE [23] calculate
Metric Formula Range Desirable
three types of errors in implemented models, so the low-
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜n 𝑇𝑃/ 𝑇𝑃 +𝐹𝑃 0-100% Max
est value is considered for an ideal model. MAE in math-
Sensitivity/ Recall 𝑇𝑃 /𝑇𝑃 + 𝐹𝑁 0-100% Max ematics is the arithmetic equivalent of absolute errors.
Specificity 𝑇𝑁 /𝑇𝑁 + 𝐹𝑃 0-100% Max This criterion only measures the magnitude of the error
Accuracy 𝑇𝑃 + 𝑇𝑁/𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 0-100% Max and does not give a significant indication of the direction
+ 𝐹𝑁 of the error. These three criteria have been used in a study
𝐹 − 𝑀𝑒𝑎𝑠𝑢𝑟e (2 * Precision * Recall) / 0–1 Max to evaluate the model’s performance [23, 33].
(Precision + Recall) The Kappa statistic is a tool that evaluates the effective-
Positive Predictive (sensitivity * prevalence) / 0-100% Max ness of a model’s reliability and ensures an accurate rep-
Value (PPV) [ (sensitivity * prevalence) resentation of changes in collected data. It can range
+ ((1 – specificity) * (1
– prevalence))
from − 1 to + 1. Despite being a common statistic, there
Kappa statistic 2*(𝑇𝑃 *𝑇𝑁- 𝐹𝑁* 𝐹𝑃)/ 0–1 Max
is no consensus on measures of it in health studies. It has
(𝑇𝑃+ 𝐹𝑃)*(𝐹𝑃+ 𝑇𝑁)*(𝑇𝑃+ been utilized solely in one study [38].
𝐹𝑁)*(𝐹𝑁+ 𝑇𝑁) Most studies commonly used Sensitivity [2, 18, 28–30,
Mean Absolute (1/n) Σ(i = 1 to n) |y_i – ŷ_i| LB:0 Min 32–37], Specificity [2, 28], and Accuracy [2, 18, 26, 28–
Error (MAE) 30, 32–37] as criteria. These metrics measures by True
Root Mean √ Σ(i = 1 to n) (y_i – ŷ_i)2/𝑁 LB:0 Min Positive (TP), True Negative (TN), False Positive (FP),
Squared Error
and False Negative (FN). F1 score or 𝐹 − 𝑀𝑒𝑎𝑠𝑢𝑟e as a
(RMSE)
harmonic average of accuracy and recall. Where there is
Root Relative √Σ(i = 1 to n) (ti − ri)2 /Σ(i = 1 0-∞ Min
Squared Error to n) (ti−ˉt)2 a balance between accuracy and recall, it is a better met-
(RRSE) ric [2, 18, 26, 28–30, 32–37].
Area Under the 0–1 Max
Curve (AUC) Specifications of decision support systems for PCOS
Precision-Recall 0–1 Max prediction
Curve (AUPR) As presented in Table 1, Eight studies [18, 21–23, 27, 28,
31, 34] designed a model to predict disease. These studies
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 23 of 26

used Decision trees [22, 31], Topology-preserving for- data in one study [26], the unwillingness of patients to
ward network [21], multi-layer perceptron [21], NFRS disclose reports and clinical data in one [19] and the need
[23], Artificial neural network [23], Apriori algorithm to improve accuracy using other classifiers is mentioned
[27], NB classifier method [28], LR [18, 28], KNN [18, 28, in one study [2].
31], CART [28], RF Classifier [18, 28], Gaussian Naive
Bayes [18], Fuzzy [31], CNN [34], GCN techniques [34] Quality assessment of included studies
and SVM [18, 28, 31]. The quality assessment of the included studies is detailed
in the Supplementary file (Table S4), with 17 studies
Specifications of decision support systems for PCOS rated as high quality and three as moderate qualities.
diagnosis
Automated detection models based on database: Eight Discussion
studies [2, 18, 28–30, 32, 36, 37] have implemented auto- The purpose of this study was to conduct a thorough
mated detection models based on a database to diagnose review of recommendation systems for women with
or screen patients automatically. These studies aimed to PCOS. Specifically, we focused on models or applica-
diagnose diseases automatically by creating an AI-based tions that utilized artificial intelligence. We collected
model using readily available data or data from those who information from various sources such as publication
seek treatment at health centers. The studies all followed year, country, journal or conference, sample size, age of
a similar methodology. They first collected data from participants, limitations and challenges, and results. Dur-
healthy and sick individuals. After that, they performed ing systematic review, we found five studies that shared
pre-processing to identify parameters and characteris- a similar approach [2, 11, 12, 14, 17, 19]. . We found rea-
tics. They designed the model using selected techniques sons for using mobile or AI-based recommender systems
and evaluated its performance using model evaluation in PCOS disease management. We will now delve into
metrics. The studies utilized various methods such as LR the study’s findings and other studies.
[2, 28–30], Bayesian classifier [2], DT [29, 37], SVM [18, Abhari et al. investigated nutritional recommendation
28–30, 37], CNN [32], KNN [18, 28], quadratic discrim- systems without considering a specific disease. In this
inant classifier [18, 29], RF [18, 28–30], CART [28, 30], study, we reviewed the proposed recommender systems
Gaussian naive Bayes [18], and K-means clustering [36] for polycystic ovary syndrome with its various applica-
to develop the automated diagnosis model. tions [14].
Classification models based on images: Four studies Based on the study results, obesity in people with
[24–26, 35] have developed a diagnostic model for this PCOS, with the escalation of symptoms, increases the
disease using ultrasound image classification. Ultrasound cost of treatment and reduces it, especially in infertility.
images were used to train and test the model., and differ- The expenditure of lifestyle modification with the help of
ent methods such as CNN [24–26], and SVM [26] were health and weight loss is lower than drug therapy. Modi-
used in these studies. fying lifestyle and nutrition using mobile phones is con-
Follicle segmentation models: Two studies [16, 33] cre- sidered a low-cost intervention with a lower percentage
ated a model to segment follicles in ultrasound images for of invasion [20]. The results of a 2018 study by Jacque-
automatic disease diagnosis. The model specifically diag- line A. Boyle and colleagues in Australia showed that a
noses through follicle segmentation, reducing the time quality disease management application met the needs of
needed for follicle counting. The process began with the patients; however, none of the applications reviewed had
publication of images, followed by image processing to quality [12].
segment the follicles. After extracting features, classifiers As mentioned, early diagnosis of the disease in the
were used to design the model. Techniques used during early stages is associated with risk reduction of disease
the pre-processing stage included histogram equaliza- consequences. Therefore, we may need recommender
tion, contrast enhancement, and the Wiener filter for systems to reduce risk reduction and time of the diag-
noise reduction of the images. The segmentation stage nosis and increase accuracy [32]. One study by Naila
utilized the Fuzzy logicis, Hybrid Intelligent, Water Drop Jan and colleagues in 2023 investigated AI techniques
(IWD), KNN, SVM [33] and K-means clustering [16]. for PCOS diagnosis. The results of this study show that
early diagnosis of this disease is difficult despite different
Limitations and challenges mentioned in the studies symptoms in people, so automatic detection systems can
As presented in Tables 1 and 10 studies pointed out the be used as an accurate solution in this field [17]. As with
limitations and refinements. Accordingly, the small vol- our results, the existence of a limited amount of data is
ume of sample size and features in 4 studies, the need considered as one of the obstacles to the implementation
to conduct more studies in 3 studies, the increase in the of this type of study [17, 27].
time of automatic diagnosis by the system by using more
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 24 of 26

Two studies conducted by Naila Jan and colleagues in providing nutrition-based treatment solutions is signifi-
India in 2022 reported a PCOS rate of 3.7 to 22.5%, which cant. Therefore, we recommend that countries, particu-
was higher in urban than rural women [11, 17]. Based larly those with a high number of affected individuals,
on the review, 40.9% of studies have been conducted prioritize policies that encourage further studies to eval-
in India, which justifies the high rate of PCOS. Also, uate the effectiveness of recommendation systems on the
unhealthy lifestyles, including unhealthy eating, can be a nutrition of people with polycystic ovary syndrome. We
reason for most urban women to do this. can work on the quality of life for individuals affected by
Among the models designed to predict PCOS, the best this condition.
accuracy belongs to a fuzzy logic-based model with an
Abbreviations
accuracy of 98.2% [31]. In the field of PCOS diagnosis, a AI Artificial Intelligence
hybrid model based on ANN, CNN and InceptionV3 has PCOS Polycystic ovary syndrome
the best performance among the designed models with SOM Self-Organizing Map
TPFN Topology-Preserving Forward Network
an accuracy of 98.12 [35]. RF and SVM were two widely MLP Multi-Layer Perceptron
used algorithms with acceptable performance, but the NFRS Neural Fuzzy Rough Set Evaluation
performance of the CNN-based model with 97% accu- ANN Artificial Neural Network
KNN K-Nearest Neighbor
racy is better than these two algorithms [32]. Based on CART Classification and Regression Trees
the study by Naila Jan and colleagues, the best perfor- SVM Support Vector Machine
mance belonged to a CNN-based system with an average LR Logistic Regression
RF Random Forest
of 76.36% and a micro-average f1-score of 100%. KNN, SFFS Sequential Forward Floating Selection
ANN, and Fuzzy logic with an accuracy of 97%, 97.5%, DT Decision Tree
and 97.30 were the best classification techniques among RCT Clinical Randomized Trial
NB Naïve Bayes
the reviewed articles [11, 17]. GCN Graph convolutional network
Another study in 2018 by Jue Xie and colleagues in CNN Convolutional Neural Networks
Australia aimed to investigate the AskPCOS application LDR Linear Decision Analysis

and the steps involved in creating it. The results showed

that the mentioned program is one of the best evidence- Supplementary Information
based user programs to manage PCOS disease. Due to The online version contains supplementary material available at https://doi.
org/10.1186/s12905-024-03074-3.
the support of 5 common languages of the world, it can
eliminate the inequality of lack of access to information Supplementary Material 1
in developing countries. According to the evaluation, 80%
of people were satisfied. The application’s usefulness was Acknowledgements
70%. Jacqueline A. Boyle’s study showed that evidence- We would like to express our gratitude to the Tehran University of Medical
based application programs positively affect the patient’s Sciences for their invaluable support.
awareness [19]. Author contributions
ABY and FS collected the data and wrote the first manuscript. LSH, LA, ShR.NK
Study limitations The study has several limitations. (1) and JF supervised all stages of the study. All authors reviewed and approved
the final manuscript.
only two studies addressed the aspect of concomitant
health management in PCOS. (2) there is a lack of access Funding
to several paid articles that can provide better results by No funding sources.
reviewing them. 3, the case was a similar study in other Data availability
developing countries such as Iran, which may cause prob- The datasets generated and/or analysed during the current study are available
lems in the results of this study in this country. Therefore, in the PubMed, Scopus, and Web of Science databases.
we recommend carrying out more specialized studies
about recommender systems. We should use AI to modify Declarations
lifestyles to nutritional patterns, such as providing a dedi- Competing interests
cated budget, creating motivation, and creating a culture The authors declare no competing interests.
among researchers.
Ethics approval and consent to participate
This study was carried out in the first phase of the thesis entitled “Design and
Conclusion evaluation of the application program of nutrition advisor for women with
We conducted a systematic review to explore the use of polycystic ovary syndrome (PCOS)” and was approved by the Research Ethics
Committee of Tehran University of Medical Sciences, Tehran, Iran (ethical code:
AI and companion health systems in managing polycys- IR.TUMS.SPH.REC.1402.198).
tic ovary syndrome, with a focus on nutrition. Although
AI has primarily been used for disease diagnosis, the Consent for publication
Not applicable.
positive impact of AI and companion health systems in
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 25 of 26

Author details 16. Mandal A, Saha D, Sarkar M, editors. Follicle Segmentation Using K-Means
1
Health Information Management and Medical Informatics Department, Clustering from Ultrasound Image of Ovary. Proceedings of International
School of Allied Medical Sciences, Tehran University of Medical Sciences, Conference on Frontiers in Computing and Systems; 2021 2021//; Singapore:
Tehran, Iran Springer Singapore.
2
Department of Community Nutrition, School of Nutritional Sciences and 17. Makhdoomi A, Jan N, Palak P, Goel N, editors. Machine learning techniques
Dietetics, Tehran University of Medical Sciences, Tehran, Iran for medical images in PCOS. 2022 4th International Conference on Artificial
3
Endocrinology and Metabolism Research Centre, Tehran University of Intelligence and Speech Technology (AIST); 2022 9–10 Dec. 2022.
Medical Sciences, Tehran, Iran 18. Vedpathak S, Thakre V, PCOcare:. PCOS detection and prediction using
4
Health Information Technology Department, School of Allied Medical machine learning algorithms. Biosci Biotechnol Res Commun. 2020;13:240–4.
Sciences, Zabol University of Medical Sciences, Zabol, Balouchestan, 19. Xie J, Burstein F, Garad R, Teede HJ, Boyle JA. Personalized Mobile Tool AskP-
Sistan, Iran COS delivering evidence-based Quality Information about polycystic ovary
5
Students’ Scientific Research Center (SSRC), Tehran University of Medical syndrome. Semin Reprod Med. 2018;36(1):66–72.
Sciences, Tehran, Iran 20. Jiskoot G, Benneheij SH, Beerthuizen A, de Niet JE, de Klerk C, Timman R, et al.
6
Department of Health Information Technology, School of Paramedical A three-component cognitive behavioural lifestyle program for preconcep-
and Rehabilitation Sciences, Mashhad University of Medical Sciences, tional weight-loss in women with polycystic ovary syndrome (PCOS): a proto-
Mashhad, Iran col for a randomized controlled trial. Reproductive Health. 2017;14(1):34.
7
Student Research Committee, Saveh University of Medical Sciences, 21. Lehtinen J-C, Forsström J, Koskinen P, Penttilä T-A, Järvi T, Anttila L. Visualiza-
Saveh, Iran tion of clinical data with neural networks, Case study: polycystic ovary
syndrome. Int J Med Informatics. 1997;44(2):145–55.
Received: 12 January 2024 / Accepted: 4 April 2024 22. Zhang H, Legro RS, Zhang J, Zhang L, Chen X, Huang H, et al. Decision
trees for identifying predictors of treatment effectiveness in clinical trials
and its application to ovulation in a study of women with polycystic ovary
syndrome. Hum Reprod (Oxford England). 2010;25(10):2612–21.
23. Rethinavalli S, Manimekalai M. A novel hybrid framework for risk severity of
References polycystic ovarian syndrome. 2016;9:19–27.
1. Akre S, Sharma K, Chakole S, Wanjari MB. Recent advances in the 24. Cahyono B, Adiwijaya, Mubarok MS, Wisesty UN, editors. An implementation
management of polycystic ovary syndrome: a review article. Cureus. of convolutional neural network on PCO classification based on ultrasound
2022;14(8):e27689. image. 2017 5th International Conference on Information and Communica-
2. Mehrotra P, Chatterjee J, Chakraborty C, Ghoshdastidar B, Ghoshdastidar S, tion Technology (ICoIC7); 2017 17–19 May 2017.
editors. Automated screening of Polycystic Ovary Syndrome using machine 25. Dewi RM, Adiwijaya, Wisesty UN, Jondri. Classification of polycystic ovary
learning techniques. 2011 Annual IEEE India Conference; 2011 16–18 Dec. based on ultrasound images using competitive neural network. J Phys: Conf
2011. Ser. 2018;971(1):012005.
3. Nilashi M, Ahmadi H, Shahmoradi L, Mardani A, Ibrahim O, Yadegaridehkordi 26. Thufailah IF, Adiwijaya, Wisesty UN, Jondri. An implementation of Elman neu-
EJJSC et al. Knowledge Discovery and diseases Prediction: a comparative ral network for polycystic ovary classification based on ultrasound images. J
study of machine. Learn Techniques. 2017;4(5). Phys: Conf Ser. 2018;971(1):012016.
4. Alotaibi M, Alsinan A, editors. A mobile Polycystic ovarian syndrome manage- 27. Boddu V, Bollineni A, Bhargav K, Sarangi S, Chilla M. Application of the Apriori
ment and awareness system for Gulf countries: System architecture. 2016 SAI Algorithm for Prediction of polycystic ovarian syndrome (PCOS). 2018. p.
Computing Conference (SAI); 2016 13–15 July 2016. 934–44.
5. Shahmoradi L, Safdari R, Ahmadi H, Zahmatkeshan M. Clinical decision 28. Denny A, Raj A, Ashok A, Ram CM, George R, editors. i-HOPE: Detection And
support systems-based interventions to improve medication outcomes: a Prediction System For Polycystic Ovary Syndrome (PCOS) Using Machine
systematic literature review on features and effects. Med J Islamic Repub Iran. Learning Techniques. TENCON 2019–2019 IEEE Region 10 Conference (TEN-
2021;35:27. CON); 2019 17–20 Oct. 2019.
6. Shahmoradi L, Rezayi SJJoN. Rehabilitation. Cognitive rehabilitation in people 29. Adla YAA, Raydan DG, Charaf MZJ, Saad RA, Nasreddine J, Diab MO, editors.
with autism spectrum disorder: a systematic review of emerging virtual Automated Detection of Polycystic Ovary Syndrome Using Machine Learning
reality-based approaches. 2022;19(1):91. Techniques. 2021 Sixth International Conference on Advances in Biomedical
7. Maghooli K, Langarizadeh M, Shahmoradi L, Habibi-Koolaee M, Jebraeily Engineering (ICABME); 2021 7–9 Oct. 2021.
M, Bouraghi HJAIM. Differential diagnosis of erythmato-squamous diseases 30. Hassan M, Mirza T. Comparative Analysis of Machine Learning Algorithms in
using classification and regression tree. 2016;24(5):338. diagnosis of polycystic ovarian syndrome. Int J Comput Appl. 2020;Volume
8. Gopal G, Suter-Crazzolara C, Toldo L, Eberhardt WJCC, Medicine L. Digital 175.
transformation in healthcare–architectures of present and future information 31. Kodipalli A, Devi S. Prediction of PCOS and Mental Health using fuzzy infer-
technologies. 2019;57(3):328–35. ence and SVM. Front Public Health. 2021;9:789569.
9. Nilashi M, Ahmadi N, Samad S, Shahmoradi L, Ahmadi H, Ibrahim O, et al. 32. Lv W, Song Y, Fu R, Lin X, Su Y, Jin X, et al. Deep Learning Algorithm for Auto-
Disease Diagnosis Using Mach Learn Techniques: Rev Classif. 2020;7(1):19–30. mated Detection of Polycystic Ovary Syndrome using scleral images. Front
10. Shahmoradi L, Almasi S, Ahmadi H, Bashiri A, Azadi T, Mirbagherie A, et al. Endocrinol. 2021;12:789878.
Virtual reality games for rehabilitation of upper extremities in stroke patients. 33. N.S N, Novel Hybrid RDA. Algorithm with Extreme Learning adaptive neuro-
J Bodyw Mov Ther. 2021;26:113–22. fuzzy inference system based follicle identification for polycystic ovary
11. Jan N, Makhdoomi A, Handa P, Goel N, editors. Machine learning approaches syndrome in Ultrasound images. Webology. 2021;18:594–616.
in medical image analysis of PCOS. 2022 International Conference on 34. Zhang N, Wang H, Xu C, Zhang L, Zang T. DeepGP: an Integrated Deep Learn-
Machine Learning, Computer Systems and Security (MLCSS); 2022: IEEE. ing Method for Endocrine Disease Gene Prediction using. Omics Data. 2021;9.
12. Boyle JA, Xu R, Gilbert E, Kuczynska-Burggraf M, Tan B, Teede H, et al. Ask 35. Hosain AS, Mehedi MHK, Kabir IE, editors. Pconet: A convolutional neural
PCOS: identifying need to inform evidence-based App Development for network architecture to detect polycystic ovary syndrome (pcos) from ovar-
Polycystic Ovary Syndrome. Semin Reprod Med. 2018;36(1):59–65. ian ultrasound images. 2022 International Conference on Engineering and
13. Portugal I, Alencar P, Cowan D. The use of machine learning algorithms in rec- Emerging Technologies (ICEET); 2022: IEEE.
ommender systems: a systematic review. Expert Syst Appl. 2018;97:205–27. 36. Zigarelli A, Jia Z, Lee H. Machine-aided self-diagnostic prediction models
14. Abhari S, Safdari R, Azadbakht L, Lankarani KB, Niakan Kalhori SR, Honarvar B, for polycystic ovary syndrome: Observational Study. JMIR Formative Res.
et al. A systematic review of Nutrition Recommendation systems: with Focus 2022;6(3):e29967.
on Technical Aspects. J Biomedical Phys Eng. 2019;9(6):591–602.
15. Kaur R, Kumar R, Gupta M. Food Image-based diet recommendation frame-
work to overcome PCOS problem in women using deep convolutional neural
network. Comput Electr Eng. 2022;103:108298.
Shahmoradi et al. BMC Women's Health (2024) 24:234 Page 26 of 26

37. Nsugbe E. An artificial intelligence-based decision support system for early

diagnosis of polycystic ovaries syndrome. Healthc Anal. 2023;3:100164. Publisher’s Note
38. McHugh ML. Interrater reliability: the kappa statistic. Biochemia Med. Springer Nature remains neutral with regard to jurisdictional claims in
2012;22(3):276–82. published maps and institutional affiliations.