Page 11 - GTM-2-2

Global Translational Medicine Clinical algorithms in ART

factors, to predict the likelihood of a successful response (2PNs), and usable blastocysts. This model can potentially
to COS. These algorithms can then be used to inform improve outcomes for many IVF patients (Table 2). After
treatment decisions, such as selecting the appropriate dose providing input information about the total amount of
of ovarian stimulation medication (Table 2). By adjusting follicles binned by size (<11 mm, 11–13 mm, 14–15 mm,
treatment protocols based on these algorithms, clinicians 16–17 mm, 18–19 mm, and >19 mm) and the estradiol level
can optimize outcomes while minimizing risks such as on a given examination day, the number of MII oocytes is
ovarian hyperstimulation syndrome (OHSS). For example, predicted through two different linear regression models
the follicle-stimulating hormone (FSH) dosing algorithm corresponding to two distinct scenarios hypothesizing
has been used to predict response to COS with progressive triggering the same day and triggering the day after,
improvement using large datasets of patient characteristics. respectively. If the predicted number of MII oocytes “today
Another approach that has been used to optimize versus tomorrow” shows a decreasing trend, triggering is
COS treatment protocols is reinforcement learning. recommended; otherwise, if the number of MII oocytes
Reinforcement learning algorithms can learn to optimize is expected to be higher if triggering the day after, it
treatment protocols by iteratively adjusting treatment could be worth waiting one more day before updating
parameters based on feedback from previous patients. For the follicle count and the estradiol level and repeating the
example, an algorithm could learn to adjust the dose and previous step. A third linear regression model can also be
timing of stimulation medication to maximize the number used to predict the estradiol level 1 day after. The trigger
of eggs retrieved while minimizing side effects. calculation to optimize trigger time and oocyte retrieval is
a strong advantage in clinical practice.
Real-time monitoring algorithms can be used to develop
personalized treatment plans that optimize outcomes 3.8. The number of oocytes exposed to fertilization
while minimizing risks. For example, a deep learning The number of oocytes that should be exposed to
algorithm could analyze ultrasound images of ovarian fertilization during an ART cycle needs to be decided to
follicles to predict the number and quality of oocytes that minimize the number of unused embryos and optimize
will be retrieved. By integrating these predictions into the probability of live birth. A tool for prediction was
treatment protocols, clinicians can optimize outcomes developed during a study on IVF cycles, which can assist
while minimizing risks. clinicians in determining the most suitable number of

These algorithms can optimize COS in ART, improving oocytes to be exposed to sperm. This can help reduce
the efficiency and effectiveness of treatment while the number of unused embryos generated and effectively
minimizing risks and improving patient outcomes [26,27] address any existing concerns of both the patients and
(Table 2). clinicians involved (Table 2). The optimization of the
[31]
number of oocytes exposed to sperm and the number of
3.6. Starting dose of gonadotropins unused embryos represent a concrete improvement of the
A recent development involves the creation of machine IVF procedure.
learning models that are interpretable and designed to
optimize the selection of starting gonadotropin doses 3.9. ART calculator
based on criteria such as mature oocytes (metaphase II In IVF/intracytoplasmic sperm injection (ICSI), an ART
[MII]), fertilized oocytes (2 pronuclear [2PN]), and viable calculator has been developed to estimate the minimum
blastocysts [28,29] . Fanton et al. have proposed a machine number of MII oocytes required for obtaining at least one
learning model for selecting the initial FSH that can euploid blastocytes for each patient, serving as a useful tool
deliver optimal laboratory results while minimizing the for counseling and planning treatments [31,32] . This prediction
use of starting and total FSH . Another machine learning tool is highly beneficial for clinical and embryological daily
[30]
model has been introduced by Correa et al. as a training practice. In addition, Jin et al. have created a nomogram
and educational resource for new clinicians and as a means to predict blastocyst formation rates based on the range of
of quality control for experienced clinicians. This model is clinical characteristics in patients with different types of
helpful in the adequate calibration of the personalization infertility, aiming to minimize the possibility of wasting
of the treatment to obtain the best number of oocytes and embryos and accurately predict the likelihood of blastocyst
[28]
avoid OHSS (Table 2). formation .The patients were categorized into three groups:
[9]
tubal factor, polycystic ovary syndrome, and endometriosis,
3.7. Optimal day of the trigger with each group further divided into a training set and a
The model suggested by Fanton et al. optimize the day validation set. The nomogram was constructed using the
[30]
of trigger for mature oocytes (MII), fertilized oocytes training set, while the performance of the model was tested

Volume 2 Issue 2 (2023) 5 https://doi.org/10.36922/gtm.0308