Available online at [Link].

com Current Opinion in

ScienceDirect Endocrine and Metabolic Research

Reviews

Mathematical modelling of follicular growth and ovarian

stimulation
Sophie Fischer-Holzhausen and Susanna Röblitz

Abstract female or/and male reproductive system [13]. Female

The aim of ovarian stimulation in fertility treatment is to in- factors, such as ovulatory disorders, endometriosis and
crease the number of large follicles and hence the number of tubal abnormalities, are responsible for approximately
eggs that can be retrieved for in vitro fertilisation (IVF). How- one-third of all cases [62].
ever, large inter- and intra-individual variability in the menstrual
cycle and ovarian response to stimulation drugs complicate The hypothalamic-pituitary-gonadal (HPG) axis is
treatment planning and prediction. Hence, many mathematical central to enable reproduction in both sexes. In females,
models have been developed to support treatment decisions. the HPG axis regulates the menstrual cycle, including
In this article, we give an overview of mechanistic models that the maturation and release of oocytes, the periodic
cover different aspects of the processes involved in normal release of reproductive hormones, as well as the prepa-
menstrual cycles and ovarian stimulation, including hormonal ration of the female body for a possible pregnancy.
regulation and follicular maturation. We also review statistical This is enabled through the feedback interactions be-
models that have been designed to predict different IVF tween ovarian hormones, mainly progesterone (P4) and
outcome criteria. Finally, we outline the use of mathematical oestradiol (E2), the pituitary hormones luteinising
models for in-silico clinical trials in reproductive endocrinology. hormone (LH) and follicle-stimulating hormone (FSH),
and the hypothalamic hormone gonadotropin-releasing
Addresses hormone (GnRH), see Fig. 1. GnRH stimulates the
University of Bergen, Computational Biology Unit, Department of
release of FSH and LH. Both regulate follicular matu-
Informatics, Thormøhlensgate 55, Bergen, 5008, Norway
ration [15]. However, the initial recruitment of follicles
Corresponding author: Röblitz, Susanna ([Link]@[Link]) from the ovarian reservoir is independent of LH and
FSH [44]. Within each menstrual cycle cohorts of fol-
licles, called waves, start growing as a result of increasing
FSH levels, see Fig. 2 [4e6]. Growing follicles produce
Current Opinion in Endocrine and Metabolic Research 2022,
26:100385
E2, which enables a feedback loop back to the hypo-
thalamus. Usually one follicle of the cohort, rarely
This review comes from a themed issue on Mathematical Modelling of
Endocrine Systems
multiple follicles, ovulates around mid-cycle. During
ovulation, the follicle releases its oocyte, and the sac
Edited by Craig McArdle, Krasimira Tsaneva-Atanasova and Mar-
garitis Voliotis
forms the corpus luteum, which produces ovarian hor-
mones in the luteal phase. If the oocyte is not fertilised
For complete overview of the section, please refer the article collection -
Mathematical Modelling of Endocrine Systems
and pregnancy does not occur, the corpus luteum decays
and a new cycle begins [15]. Failure in this endocrine
Available online 6 August 2022
network is one cause of infertility.
[Link]
2451-9650/© 2022 The Author(s). Published by Elsevier Ltd. This is an Treatment options for infertility depend on its cause
open access article under the CC BY license ([Link] and the patient itself. In vitro fertilisation (IVF) is a
org/licenses/by/4.0/).
form of assisted reproductive technology (ART) that can
result in a successful pregnancy for patients suffering
Keywords from different causes of infertility. IVF can not only be
Hypothalamic-pituitary-gonadal (HPG) axis, Hormone dynamics,
used to overcome female infertility but also assist in
Follicular competition, Assisted reproductive technology (ART).
cases of male infertility. Intracytoplasmic sperm injec-
tion (ICSI), a technique where a sperm cell is injected
Introduction directly into the egg cell, may be used for patients with
Approximately 15% of people of reproductive age are low sperm quality or number. IVF treatment proceeds
affected by infertility [67], and unwanted childlessness in three steps: (i) egg retrieval through controlled
puts a psychological and psycho-social burden on many ovarian stimulation (COS), (ii) fertilisation of oocytes in
of them [27,41,14]. This makes infertility and its con- the laboratory and (iii) embryo transfer into the uterus
sequences a global health issue. In 85% of all cases of [2]. COS aims to stimulate the growth of multiple
infertility, the underlying causes are dysfunctions in the ovarian follicles synchronously by the administration of

[Link] Current Opinion in Endocrine and Metabolic Research 2022, 26:100385

2 Mathematical Modelling of Endocrine Systems

Figure 1 Figure 2

Current Opinion in Endocrine and Metabolic Research

Illustration of the follicular wave theory. Small follicles are available

throughout the menstrual cycle. With the beginning of a new menstrual
cycle, the follicle-stimulating hormone level (FSH, pink line) starts rising
and stimulates the growth of a cohort of follicles. Until mid-cycle, one
follicle will be selected for ovulation. During the luteal phase, another
cohort of follicles starts growing. However, none of those follicles will
ovulate due to the low level of FSH.

in terms of pregnancy and live birth rates while simul-

Schematic representation of the hypothalamic-pituitary-gonadal taneously decreasing treatment-related risks like ovarian
(HPG) axis. Gonadotropin-releasing hormone (GnRH) is synthesised in hyperstimulation syndrome [53].
the hypothalamus and released into the hypophyseal portal circulation
system in a pulsatile manner. In the pituitary, GnRH stimulates the syn- Mathematical modelling can improve our understanding
thesis of luteinising hormone (LH) and follicle-stimulating hormone (FSH)
and their release into the blood. LH and FSH regulate follicular maturation
of complex regulatory networks involving multiple levels
in the ovaries. FSH stimulates follicular growth, whereas LH triggers the of organisation, such as endocrine systems [73,37].
ovulation of a dominant follicle. Growing follicles (yellow) produce Computational models can also be helpful to answer
oestradiol (E2). After ovulation (dark grey), the corpus luteum (orange) scientific questions in cases where appropriate model
produces both E2 and progesterone (P4). E2 and P4 exhibit feedback organisms are not available, and experimental in-
mechanisms on the hypothalamus and the pituitary, which closes the loop
(created with [Link]).
vestigations are challenging. Until recently, a menstrual
cycle was only observed in primates. Evidence of a
menstruating rodent was provided [9]. Since every
gonadotrophins [40]. Several COS protocols are avail- model is a simplification and based on assumptions, it is
able. The conventional GnRH agonist protocol and the important to find a model that is appropriate for the
GnRH antagonist protocol are well established [30]. research question [66,65]. Mathematical models can be
Newer protocols are based on the follicular wave theory divided into two main groups: (i) empirical models and
[6], which motivates ovarian stimulation at different (ii) mechanistic models [7,51]. Empirical models are
time points within a menstrual cycle [54]. Protocols statistical models and therefore data driven. They are
which start ovarian stimulation at a random time point tailored towards prediction and are widely used in
are of particular interest in the context of fertility medical research [26]. An example in the scope of this
preservation in cancer patients, where time is a deter- review is the prediction of menstrual cycle length
mining factor [64,12]. Stimulation treatments that start [47,39]. Mechanistic models are process-based and
during the luteal phase have been used to treat patients consider the elements forming a system and their in-
who did not respond to conventional protocols [50,29] as teractions [11]. Their strength lies in generating,
well as women with normal ovarian response [33]. testing, and refining hypotheses [20]. An example is the
Double stimulation protocols comprise two consecutive model by Ref. [31] that provides evidence for the
treatment cycles and offer more opportunities for oocyte follicular wave theory.
retrieval in a shorter time interval [32,63]. Overall, it is
challenging to find the best therapy for an individual In the following, we give an overview of statistical
patient in order to achieve the best treatment outcome models (Sec. 2) and mechanistic models (Sec. 3) that

Current Opinion in Endocrine and Metabolic Research 2022, 26:100385 [Link]

 Modelling follicle growth and ovarian stimulation Fischer-Holzhausen and Röblitz 3

have been developed to simulate and predict Ref. [60] introduced a logistic regression model based
IVF treatment outcomes, including the number of on seven predictors for estimating the assisted fecundity
mature oocytes after ovarian stimulation and pregnancy of women before starting the first IVF/ICSI cycle, which
rates. In addition, we briefly review the use of mathe- translates into the probability of live birth in the first
matical models in in-silico clinical trials (ISCT) related treatment cycle. This kind of prediction complements
to IVF treatment (Sec. 4). the approach of estimating cumulative and cycle-
specific probabilities of live birth over multiple treat-
Prediction of IVF outcomes ment cycles.
A review by Ref. [56] summarises the statistical
methods and available software tools for scoring embryo A great challenge for ART is a poor ovarian response,
quality and predicting pregnancy rates. The review which refers to an unexpected low number of oocytes
emphasises the role of these models as a clinical decision upon stimulation treatment. Using univariate and
support tool, and that the final decisions need to be multivariate logistic regression analyses, Ref. [69] devel-
made by the practitioners and laboratory staff. Another oped a statistical model based on four predictors (anti-
publication from the same group [55] gives an overview Müllerian hormone, antral follicle counts, basal FSH, and
of statistical models based on patient and/or embryo age, in order of their significance) in order to estimate the
characteristics for predicting pregnancy and/or live birth probability of poor ovarian response and to assess the true
rates. The authors suggest that the way forward would ovarian reserve. Similarly, Ref. [38] developed a statistical
be in enriching, improving and strengthening the model that can predict the probability of clinical preg-
reproducibility and prognostic value of current models nancy failure in poor ovarian responders before embryo
instead of suggesting new ones. However, both the transfer in IVF/ICSI procedure.
definition of new success criteria as well the advance-
ment of computational methods and tools, particularly ML models allow for including an increased number of
in the fields of machine learning (ML) and artificial features and to untangle their complex relationships.
intelligence (AI), has led to the development of new Ref. [8] compared two widely used ML methods (sup-
models over the past years, which we briefly review in port vector machines with different kernel functions and
the following. artificial neural networks) with logistic regression models
for the prediction of different IVF outcome criteria.
An intermediate marker of successful outcome in IVF/ They demonstrated that the ML methods are superior to
ICSI cycles, which has been introduced by the the standard statistical models. The authors argue that
POSEIDON group,1 is the ability to retrieve the number ML algorithms, as opposed to classical statistical models,
of oocytes needed to achieve at least one euploid embryo can take into consideration complex associations be-
for transfer, i.e., an embryo that contains a normal tween different parameters and can consequently better
number of chromosomes. In Ref. [21], members of the utilise the synergism between these associated parame-
POSEIDON group developed a statistical model to es- ters. In the same direction, Ref. [25] used 25 attributes
timate the minimum number of mature oocytes required in combination with a feature selection algorithm to
to obtain at least one euploid blastocyst (based on pre- assess the prediction ability of IVF pregnancy success for
treatment information, including female age and sperm five different ML models. Two features, namely indica-
source used for ICSI) and to estimate the individualised tion of infertility factor and the number of mature eggs,
probability of blastocyst euploidy per mature retrieved were selected by all classifiers, and antral follicle count
oocyte. External multicentre validation is currently (AFC) was selected by four methods. Moreover, age was
ongoing using suitable ART datasets from different ranked highest by three of the classifiers, which is
countries. consistent with other studies. The authors demonstrated
that the prediction performance of all five classifiers
There are also studies that focus on predicting the total improved with the selected features compared to using
number of oocytes. Ref. [1] demonstrated that results all features. Their article also includes a summary of
from random forest analysis were consistent with a studies that applied ML techniques for the classification
generalised linear regression model suggesting that fol- of IVF outcomes. Those techniques differ in the ML
licle sizes of 12e19 mm (but not the total number of technique used, the attribute/feature selection tech-
follicles) on the day of trigger had the greatest predic- nique used, the list of selected features, the validation
tive importance for the number of oocytes and number (training/test procedure), and the performance measure
of mature oocytes retrieved. This knowledge enables reported. These differences make it difficult to compare
the accurate determination of trigger efficacy and could the methods with each other and also limit their trans-
potentially also be used to determine the optimal day of ferability to other clinics due to variations in the amount
trigger administration. and quality of data. The publication of codes as well as
the availability of benchmark datasets would be prefer-
able in order to increase the reproducibility and reus-
1
[Link] ability of ML methods and results.
[Link] Current Opinion in Endocrine and Metabolic Research 2022, 26:100385
4 Mathematical Modelling of Endocrine Systems

Modelling follicular maturation on a development of ovarian follicles and ovarian hormone

systems level production. Proliferation and cellular signalling processes
Both classical statistical models as well as ML models of these cells have been investigated experimentally and
are based on predefined input and output variables. by mathematical modelling [16]. Recently, Ref. [18]
They do not explicitly include time as a variable and can introduced a continuous-time Markov chain model for
therefore not be used to predict system behaviour over cell population dynamics to identify events in follicle
time, e.g., the growth of ovarian follicles. For this pur- maturation. Modelling follicular maturation on different
pose, process-based models have been developed, which levels of organisation, for example, by incorporating cell
will be summarised in the following. dynamics in follicle population dynamic models, can be
valuable to characterise the pool of follicles over the
Moment models that describe the response of lifetime of individuals [17,10].
follicles in IVF
Ref. [72] developed a mathematical model that de- Models based on follicular maturation stages and
scribes how the discrete follicle size distribution masses
evolves over time, whereby it is assumed that the A number of models have been developed [24,48,49,46]
number of follicles activated during an IVF cycle is in which discrete stages of follicular maturation are
constant, i.e., that no new follicles start growing during defined as a state variables to describe follicular growth
stimulation. The kinetics of follicle growth is modelled dynamics. Thereby, each maturation stage encodes a
as a function of injected FSH, and the follicle properties specific capability to produce ovarian hormones, but
are represented in terms of the moments of the (un- the variables do not refer to the size or number of folli-
known) statistical size distribution. Initial data from cles in that stage of maturation. However, this heuristic
two treatment days (follicle sizes and prescribed FSH approach is useful to study different aspects of the
dose on days 2 and 5) of an individual patient are used female menstrual cycle. For example, Refs. [49,68] used
to obtain patient-specific model parameters and predict this approach to model drug administrations, while
the follicle size distribution for the remaining treat- Ref. [46] investigated follicular wave dynamics. Ref. [23]
ment days, whereby the dose is not adjusted but con- used the model to investigate the effect of testosterone
stant throughout treatment. The authors demonstrate on normal menstrual cycles and ovulatory function. The
that the follicle size distribution predicted by the model provides a framework to investigate polycystic
moment model is in good agreement with the actual ovary syndrome and ovulatory dysfunctions.
size distribution seen in the IVF cycle data for five
patients. Follicle population models
A mathematical formulation for ovarian follicle matura-
In [71], the authors extended their model by an optimal tion dynamics in terms of number and sizes of follicles
control approach in order to predict the optimum FSH was first introduced to the literature by Refs. [34,35].
dosage for the desired treatment outcome, which is to Ref. [52] used this model to predict ovarian response in
have as many follicles as possible in the largest size class. stimulation treatments. Ref. [59] modified the Lacker’s
A proof of concept based on data from five patients was model in order to simulate higher ovulations rates, i.e.,
presented in Ref. [71], before the model was tested in a double and multiple ovulations, in sheep and cattle.
double-blinded trial involving 10 patients [45]. Even Based on these previous modelling attempts for follic-
though the cohort size was small, the results from ular maturation on the level on individual follicles,
Ref. [45] demonstrate that model-based treatment Ref. [36] introduced a follicular growth equation that
planning can lead to lower doses and fewer tests and includes competition between follicles, with the follic-
monitoring requirements along with higher numbers of ular size as state variable. All these models, however, can
mature follicles and a similar percentage of good quality only be used to simulate one follicular wave. Ref. [22]
eggs compared to standard treatment routines. coupled the Lange model with the hormone dynamics
along the HGP axes. The coupled model allows us to
Since the model does not include hormone dynamics study the interplay between hormone dynamics and
and does not consider outcomes other than follicle follicular maturation throughout consecutive menstrual
number and sizes, the risk of ovarian hyperstimulation cycles and can be used to simulate ovarian stimulation
syndrome still needs to be checked by the physician, protocols with random start times.
which might overrule the model-based treatment sug-
gestions in many cases. Treatment computations and in-silico
clinical trials
Cellular population models Mechanistic models can be used as a safe and efficient
Ovarian follicles carry two types of hormone-sensitive tool to predict patient-specific treatment outcomes as
cells: (i) LH-responsive theca cells and FSH-responsive part of ISCT. Those approaches promise to decrease
granulosa cells. Both cell types are crucial for the experimental efforts, including animal and human

Current Opinion in Endocrine and Metabolic Research 2022, 26:100385 [Link]

 Modelling follicle growth and ovarian stimulation Fischer-Holzhausen and Röblitz 5

Figure 3

Current Opinion in Endocrine and Metabolic Research

This review gives an overview of different mathematical modelling approaches focusing on ovarian follicle maturation and female health. There are two
main model types, namely statistical models and mechanistic models, both branching into sub-classes depending on the application. Each sub-class links
to one of its most recent references, which are also cited in this review.

testing, and optimise the individual treatment outcome. stimulation in humans, see Fig. 3. It demonstrates how
The group of E. Tronci developed methods and software medical research in the context of female health already
based on intelligent search strategies, and statistical has or might in the future benefit from computational
model checking to find sets of model parameters that work, such as statistical and mechanistic modelling.
result in physiologically meaningful model behaviours Statistical models are a powerful tool to predict different
[61,43]. In Ref. [57], they applied these methods to outcome criteria of IVF treatment based on both patient
compute huge populations of virtual patients (VPs) for a and embryo characteristics. In particular, ML models
non-identifiable quantitative virtual physiological human help determine which phenotype and cycle factors are
(VPH) model of the human menstrual cycle, including the most useful in making predictions.
drug treatments [49]. Using the same VPH model,
Refs. [42,58] showcased how VPs can be used to support Mechanistic modelling, with its way of thinking about
precision medicine. Their work demonstrates how to complex dynamical systems in biology, can provide
compute a personalised down-regulation treatment valuable insights on its own [20]. In particular, mecha-
protocol (a protocol used for assisted reproduction) that nistic models can be used to test the hypothesis about
maximises the aimed outcome while simultaneously the underlying processes and identify parameters on
minimising the risk for severe side effects. which measurement efforts should be focused on.
Moreover, they can be combined with pharmacokinetic
These methods and software tools have reached a high models to study drug administration schemes, which is
level of technological readiness, and the indispensable not possible with statistical models. Recent publications
next step would be to test their performance in clinical have demonstrated how mechanistic and ML models
trials. In particular, ethical and legal issues need to be can be combined to infer hidden dynamics in biological
considered carefully before such tools can become part networks and enable robust predictions, e.g., Ref. [70].
of clinical practice [19]. This is certainly a promising avenue for future research.

Conclusion The review here focuses on follicular dynamics, and

This review summarises different mathematical ap- there are several ongoing modelling efforts in closely
proaches to model follicular maturation and ovarian related areas, for example, on the endometrial cycle

[Link] Current Opinion in Endocrine and Metabolic Research 2022, 26:100385

6 Mathematical Modelling of Endocrine Systems

[3]. Also, we did not discuss modelling approaches 12. Cakmak H, Katz A, Cedars MI, Rosen MP: Effective method for
emergency fertility preservation: random-start controlled
based on images, as this would be out of the scope for ovarian stimulation. Fertil Steril 2013, 100:1673. 1618.
this review. The reader interested in the application of
13. Carson SA, Kallen AN: Diagnosis and management of infer-
ML methods to predict embryo ploidy from images is tility: a review. JAMA 2021, 326:65–76.
referred to Ref. [28] and references therein. It is likely 14. Chow KM, Cheung MC, Cheung IK: Psychosocial interventions
that in future, different models and model types will for infertile couples: a critical review. J Clin Nurs 2016, 25:
be combined in order to achieve an even more holistic 2101–2113.
picture of the processes that are involved in 15. Christensen A, Bentley G, Cabrera R, Ortega HH, Perfito N, Wu T,
female fertility. Micevych P: Hormonal regulation of female reproduction.
Horm Metab Res 2012, 44:587–591.
16. Clément F, Crépieux P, Yvinec R, Monniaux D: Mathematical
Declaration of competing interest modeling approaches of cellular endocrinology within the
hypothalamo-pituitary-gonadal axis. Mol Cell Endocrinol 2020,
Nothing declared 518, 110877.
17. Clément F, Monniaux D: Mathematical modeling of ovarian
Acknowledgement * follicle development: a population dynamics viewpoint. Cur-
The work of SF and SR was supported by the Trond Mohn Foundation rent Opinion in Endocrine and Metabolic Research 2021.
(BSF, [Link] Grant no. BFS2017TMT01. The This review article gives an overview of modelling follicle population
funder had no role in study design, data collection and analysis, decision to dynamics on different levels of organisation and highlights the model-
publish, or preparation of the manuscript. ling of cell population dynamics in the context of folliculogenesis.
18. Clément F, Robin F, Yvinec R: Stochastic nonlinear model for
References somatic cell population dynamics during ovarian follicle
Papers of particular interest, published within the period of review, activation. J Math Biol 2021, 82:1–52.
have been highlighted as:
19. Cohen IG, Amarasingham R, Shah A, Xie B, Lo B: The legal and
* of special interest ethical concerns that arise from using complex predictive
analytics in health care. Health Aff 2014, 33:1139–1147, https://
[Link]/10.1377/hlthaff.2014.0048.
1. Abbara A, Vuong LN, Ho VNA, Clarke SA, Jeffers L,
Comninos AN, Salim R, Ho TM, Kelsey TW, Trew GH, 20. Enderling H, Wolkenhauer O: Are all models wrong? Compu-
Humaidan P, Dhillo WS: Follicle size on day of trigger most tational and Systems Oncology 2021, 1, e1008.
likely to yield a mature oocyte. Front Endocrinol 2018, 9:193.
21. Esteves SC, Carvalho JF, Bento FC, Santos J: A novel predic-
2. Anwar S, Anwar A: Infertility: a review on causes, treatment tive model to estimate the number of mature oocytes
and management. Women’s Health Gynecol 2016, 5:2. required for obtaining at least one euploid blastocyst for
transfer in couples undergoing in vitro fertilization/intra-
3. Arbeláez-Gómez D, Benavides-López S, Giraldo-Agudelo MP, cytoplasmic sperm injection: the art calculator. Front Endo-
Guzmán-Álvarez JP, Ramirez-Mazo C, Gómez-Echavarría LM: crinol 2019, 10:99.
A phenomenological-based model of the endometrial growth
and shedding during the menstrual cycle. J Theor Biol 2022, 22. Fischer S, Ehrig R, Schäfer S, Tronci E, Mancini T, Egli M, Ille F,
532, 110922. * Krüger TH, Leeners B, Röblitz S: Mathematical modeling and
simulation provides evidence for new strategies of ovarian
4. Baerwald A, Adams G, Pierson R: Characterization of ovarian stimulation. Front Endocrinol 2021, 12:117.
follicular wave dynamics in women. Biol Reprod 2003a, 69: This work introduces a mathematical model that combines hormone
1023–1031. dynamics along the HPG-axis with follicle growth dynamics on a follicle
population level in order to simulate ovarian stimulation protocols with
5. Baerwald A, Adams G, Pierson R: A new model for ovarian different start times in the cycle.
follicular development during the human menstrual cycle.
Fertil Steril 2003b, 80:116–122. 23. Graham EJ, Selgrade JF: A model of ovulatory regulation
examining the effects of insulin-mediated testosterone pro-
6. Baerwald A, Adams G, Pierson R: Ovarian antral folliculo- duction on ovulatory function. J Theor Biol 2017, 416:
genesis during the human menstrual cycle: a review. Hum 149–160.
Reprod Update 2011, 18:73–91.
24. Harris-Clark L, Schlosser P, Selgrade J: Multiple stable periodic
7. Baker RE, Peña JM, Jayamohan J, Jérusalem A: Mechanistic solutions in a model for hormonal control of the menstrual
models versus machine learning, a fight worth fighting for the cycle. Bull Math Biol 2003, 65:157–173.
biological community? Biol Lett 2018, 14, 20170660, https://
[Link]/10.1098/rsbl.2017.0660. 25. Hassan MR, Al-Insaif S, Hossain MI, Kamruzzama J: A machine
* learning approach for prediction of pregnancy outcome
8. Barnett-Itzhaki Z, Elbaz M, Butterman R, Amar D, Amitay M, following IVF treatment. Neural Comput Appl 2020, 32:
Racowsky C, Orvieto R, Hauser R, Baccarelli AA, Machtinger R: 2283–2297.
Machine learning vs. classic statistics for the prediction of The authors demonstrate that automatic feature selection combined
IVF outcomes. J Assist Reprod Genet 2020, 37:2405–2412. with ML classifiers improves the prediction performance for IVF preg-
9. Bellofiore N, Ellery SJ, Mamrot J, Walker DW, Temple-Smith P, nancy rates. The article also contains an overview of studies that
Dickinson H: First evidence of a menstruating rodent: the applied machine learning techniques for classification of IVF outcome.
spiny mouse (acomys cahirinus). Am J Obstet Gynecol 2017, 26. Henley SS, Golden RM, Kashner TM: Statistical modeling
216:40. e1. methods: challenges and strategies. Biostatistics & Epidemi-
10. Bonnet C, Chahour K, Clément F, Postel M, Yvinec R: ology 2020, 4:105–139, [Link]
Multiscale population dynamics in reproductive biology: 24709360.2019.1618653.
singular perturbation reduction in deterministic and sto- 27. Ho TTT, Le MT, Truong QV, Nguyen VQH, Cao NT: Psycho-
chastic models. ESAIM: Proceedings and Surveys 2020, 67: logical burden in couples with infertility and its association
72 – 99. with sexual dysfunction. Sex Disabil 2020, 38:123–133.
11. Brigandt I: Systems biology and the integration of mecha- 28. Huang B, Tan W, Li Z, Jin L: An artificial intelligence model
nistic explanation and mathematical explanation. Stud Hist (euploid prediction algorithm) can predict embryo ploidy
Philos Sci C Stud Hist Philos Biol Biomed Sci 2013, 44:477–492, status based on time-lapse data. BMC Reproductive Biology
[Link] and Endocrinology 2021, 19:185.

Current Opinion in Endocrine and Metabolic Research 2022, 26:100385 [Link]

 Modelling follicle growth and ovarian stimulation Fischer-Holzhausen and Röblitz 7

29. Kalra SK, Ratcliffe S, Gracia1 CR, Martino L, Coutifaris C, 48. Reinecke I, Deuflhard P: A complex mathematical model of
Barnhart KT: Randomized controlled pilot trial of luteal phase the human menstrual cycle. J Theor Biol 2007, 247:
recombinant fsh stimulation in poor responders. Reprod 303 – 330.
Biomed Online 2008, 17:745–750.
49. Röblitz S, Stötzel C, Deuflhard P, Jones HM, Azulay DO, van der
30. Karimzadeh MA, Ahmadi S, Oskouian H, Rahmani E: Compari- Graaf PH, Martin SW: A mathematical model of the human
son of mild stimulation and conventional stimulation in art menstrual cycle for the administration of GnRH analogues.
outcome. Arch Gynecol Obstet 2010, 281:741–746. J Theor Biol 2013, 321:8–27.
31. Kirillova A, Martazanova B, Mishieva N, Semenova M: Follicular 50. Rombauts L, Anne-MariaSuikkari, MacLachlan V, Trounson AO,
waves in ontogenesis and female fertility. Biosystems 2021: Healy DL: Recruitment of follicles by recombinant human
104558. follicle-stimulating hormone commencing in the luteal phase
of the ovarian cycle. Fertil Steril 1998, 69:665–669.
32. Kuang Y, Chen Q, Hong Q, Lyu Q, Ai A, Fu Y, Shoham Z: Double
stimulations during the follicular and luteal phases of poor 51. Saltelli A: A short comment on statistical versus mathematical
responders in IVF/ICSI programmes (Shanghai Protocol). modelling. Nat Commun 2019, 10:1–3.
Reprod Biomed Online 2014a, 29:684–691.
52. Sarty GE, Pierson RA: An application of Lacker’s mathemat-
33. Kuang Y, Hong Q, Chen Q, Lyu Q, Ai A, Fu Y, Shoham Z: Luteal- ical model for the prediction of ovarian response to super-
phase ovarian stimulation is feasible for producing compe- stimulation. Math Biosci 2005, 198:80–96.
tent oocytes in women undergoing in vitro fertilization/intra-
cytoplasmic sperm injection treatment, with optimal 53. Sighinolfi G, Grisendi V, La Marca A: How to personalize
pregnancy outcomes in frozen-thawed embryo transfer ovarian stimulation in clinical practice. J Turk Ger Gynecol
cycles. Fertil Steril 2014b, 101:105–111. Assoc 2017, 18:148.

34. Lacker H: Regulation of ovulation number in mammals. a 54. Sighinolfi G, Sunkara SK, La Marca A: New strategies of ovarian
follicle interaction law that controls maturation. Biophys J stimulation based on the concept of ovarian follicular waves:
1981, 35:433–454. from conventional to random and double stimulation. Reprod
Biomed Online 2018, 37:489–497.
35. Lacker HM, Akin E: How do the ovaries count? Math Biosci
1988, 90:305–332. 55. Simopoulou M, Sfakianoudis K, Antoniou N, Maziotis E, Rapani A,
Bakas P, Anifandis G, Kalampokas T, Bolaris S, Pantou A, et al.:
36. Lange A, Schwieger R, Plöntzke J, Schäfer S, Röblitz S: Follic- Making IVF more effective through the evolution of prediction
ular competition in cows: the selection of dominant follicles models: is prognosis the missing piece of the puzzle? Syst
as a synergistic effect. J Math Biol 2019, 78:579–606. Biol Reprod Med 2018a, 64:305–323.
37. Leng G, MacGregor DJ: Mathematical modelling in neuroen- 56. Simopoulou M, Sfakianoudis K, Maziotis E, Antoniou N,
docrinology. J Neuroendocrinol 2008, 20:713–718. Rapani A, Anifandis G, Bakas P, Bolaris S, Pantou A, Pantos K,
et al.: Are computational applications the “crystal ball” in
38. Li F, Lu R, Zeng C, Li X, Xue Q: Development and validation of the IVF laboratory? the evolution from mathematics to
a clinical pregnancy failure prediction model for poor ovarian artificial intelligence. J Assist Reprod Genet 2018b, 35:
responders during IVF/ICSI. Front Endocrinol 2021a, 12. 1545 – 1557.
39. Li K, Urteaga In, Shea A, Vitzthum VJ, Wiggins CH, Elhadad N: 57. Sinisi S, Alimguzhin V, Mancini T, Tronci E, Leeners B: Complete
A predictive model for next cycle start date that accounts for * populations of virtual patients for in silico clinical trials.
adherence in menstrual self-tracking. J Am Med Inf Assoc Bioinformatics 2020a, 36:5465–5472.
2021b, 29:3–11. In this article, the authors compute a population of almost five million
virtual patients for a non-identifiable quantitative virtual physiological
40. Macklon NS, Stouffer RL, Giudice LC, Fauser BC: The science
human model of the HPG axis. They demonstrate that the computed
behind 25 years of ovarian stimulation for in vitro fertilization.
parameters are physiologically meaningful, pairwise distinguishable,
Endocr Rev 2006, 27:170–207.
and complete, i.e., representative of the entire spectrum of behaviours
41. Malina A, Pooley JA: Psychological consequences of ivf defined by the given model.
fertilization–review of research. Ann Agric Environ Med 2017, 58. Sinisi S, Alimguzhin V, Mancini T, Tronci E, Mari F, Leeners B:
24:554–558. Optimal personalised treatment computation through in
silico clinical trials on patient digital twins. Fundam Inf 2020b,
42. Mancini T, Mari F, Massini A, Melatti I, Salvo I, Sinisi S, Tronci E,
174:283–310.
Ehrig R, Röblitz S, Leeners B: Computing personalised treat-
ments through in silico clinical trials. A case study on 59. Soboleva T, Peterson A, Pleasants A, McNatty K, Rhodes F:
downregulation in assisted reproduction. In Proceedings of A model of follicular development and ovulation in sheep and
25th RCRA international Workshop on experimental Evaluation of cattle. Anim Reprod Sci 2000, 58:45–57.
algorithms for Solving Problems with Combinatorial Explosion;
2018:16. 60. Tarín JJ, Pascual E, García-Pérez MA, Gómez R, Hidalgo-
Mora JJ, Cano A: A predictive model for women’s assisted
43. Mancini T, Tronci E, Salvo I, Mari F, Massini A, Melatti I: fecundity before starting the first IVF/ICSI treatment cycle.
Computing biological model parameters by parallel statistical J Assist Reprod Genet 2020, 37:171–180.
model checking. In Proceedings of the 3rd international Con-
ference on Bioinformatics and Biomedical Engineering (IWBBIO 61. Tronci E, Mancini T, Salvo I, Sinisi S, Mari F, Melatti I, Massini A,
2015). Springer; 2015:542–554. Davi F, Dierkes T, Ehrig R, Röblitz S, Leeners B, Krüger T, Egli M,
Ille F: Patient-specific models from inter-patient biological
44. McGee EA, Hsueh AJ: Initial and cyclic recruitment of ovarian models and clinical records. In Proceedings of 14th Conference
follicles. Endocr Rev 2000, 21:200–214. in formal methods in Computer-Aided design. FMCAD 2014);
2014:207–214.
45. Nisal A, Diwekar U, Bhalerao V: Personalized medicine for
* in vitro fertilization procedure using modeling and optimal 62. Unuane D, Tournaye H, Velkeniers B, Poppe K: Endocrine dis-
control. J Theor Biol 2020, 487, 110105. orders & female infertility. Best Pract Res Clin Endocrinol
Based on a moment model of follicular size distributions, the authors Metabol 2011, 25:861–873.
introduce an optimal control approach to compute patient-specific drug
doses needed to achieve an optimal ovarian stimulation outcome. 63. Vaiarelli A, Cimadomo D, Trabucco E, Vallefuoco R, Buffo L,
Dusi L, Fiorini F, Barnocchi N, Bulletti FM, Rienzi L, Ubaldi FM:
46. Panza NM, Wright AA, Selgrade JF: A delay differential equa- Double stimulation in the same ovarian cycle (DuoStim) to
tion model of follicle waves in women. J Biol Dynam 2016, 10: maximize the number of oocytes retrieved from poor prog-
200–221. nosis patients: a multicenter experience and swot analysis.
Front Endocrinol 2018, 9:317.
47. de Paula Oliveira T, Bruinvels G, Pedlar C, Moore B, Newell J:
Modelling menstrual cycle length in athletes using state- 64. von Wolff M, Thaler CJ, Frambach T, Zeeb C, Lawrenz B,
space models. Sci Rep 2021, 11, 16972. Popovici RM, Strowitzki T: Ovarian stimulation to cryopreserve

[Link] Current Opinion in Endocrine and Metabolic Research 2022, 26:100385

8 Mathematical Modelling of Endocrine Systems

fertilized oocytes in cancer patients can be started in the probability of poor ovarian response: a retrospective
luteal phase. Fertil Steril 2009, 92:1360–1365. cohort study. J Assist Reprod Genet 2020,
37:963 – 972.
65. Wolkenhauer O: Why model? Front Physiol 2014, 5:21.
70. Yazdani A, Lu L, Raissi M, Karniadakis GE: Systems biology
66. Wolkenhauer O, Mesarovi
c M: Feedback dynamics and cell informed deep learning for inferring parameters and hidden
function: why systems biology is called systems biology. Mol dynamics. PLoS Comput Biol 2020, 16, e1007575.
Biosyst 2005, 1:14–16.
71. Yenkie KM, Diwekar UM: Optimal control for predicting
67. World Health Organization: Infertility. 2022. [Link] customized drug dosage for superovulation stage of in vitro
health-topics/infertility#tab=tab_1. Accessed 4 February 2022. fertilization. J Theor Biol 2014, 335:219–228.
68. Wright AA, Fayad GN, Selgrade JF, Olufsen MS: Mechanistic 72. Yenkie KM, Diwekar UM, Bhalerao V: Modeling the superovu-
* model of hormonal contraception. PLoS Comput Biol 2020, 16, lation stage in in vitro fertilization. IEEE (Inst Electr Electron
e1007848. Eng) Trans Biomed Eng 2013, 60:3003–3008.
This work demonstrates how a model based on follicular maturation
stages coupled to the HPG-axis hormone dynamics can be used to 73. Zavala E, Wedgwood KC, Voliotis M, Tabak J, Spiga F,
simulate hormonal contraception. Lightman SL, Tsaneva-Atanasova K: Mathematical modelling of
endocrine systems. TEM (Trends Endocrinol Metab) 2019, 30:
69. Xu H, Feng G, Wang H, Han Y, Yang R, Song Y, Chen L, 244–257.
Shi L, Zhang MQ, Li R, et al.: A novel mathematical model
of true ovarian reserve assessment based on predicted

Current Opinion in Endocrine and Metabolic Research 2022, 26:100385 [Link]