Page 22 - AIH-2-3

Artificial Intelligence in Health AI in embryo selection for ART

these challenges, efforts are being made to increase the 4.5. Future scope of ART
generalizability of AI models by diversifying training Throughout the years, ART has made an extraordinary
data and developing clinic-specific models. Ethical contribution to the endeavor to resolve the infertility issues
9
considerations in adopting AI for embryo selection include that couples have been facing. ART, spanning from its
transparency, interpretability, and collaborative decision- traditional to contemporary iterations, has played a pivotal
making to ensure the well-being of prospective parents role in enhancing birth rates and facilitating conception
and uphold ethical standards in assisted reproduction. through ET, GIFT, and other procedures. It has increased
21
In addition, there is growing support for more rigorous the birth rate by surmounting barriers to conception and
clinical testing, including larger sample sizes, balanced augmenting the probability of a successful pregnancy. There
datasets, and improved performance metrics, to ensure the is significant potential for ART advancements through the
reliability and effectiveness of AI algorithms for embryo implementation of AI-based algorithms. AI possesses the
selection. capability to assist in antiretroviral therapy by addressing
71
4.4. Limitations of this study therapeutic challenges. To enhance the outcomes of ART,
prediction models can be developed by implementing ML
The studies included in this review focused on deep methodologies, which encompass a diverse array of feature
learning, ML, or AI for embryo selection, providing sets and numerous algorithms. Future advancements in
4
a general overview rather than a detailed statistical soft robotics, telesurgery, and the integration of AI with
data analysis, which could limit the depth of insights robotics may potentially lead to an ART procedure that is
provided. Furthermore, the absence of critical appraisal fully automated and intelligent. 73
may induce uncertainty regarding the robustness of the
evidence synthesized in the review. The selection was The potential impacts of AI on ARTs are shown in
determined through an exhaustive search of numerous Figure 5, highlighting seven key areas:
databases, which may have resulted in bias. Due to limited (i) Personalized treatment plans: AI customizes care
access, some databases, such as Web of Science and based on patient data, taking specific health profiles
PsycINFO, were not included in the study. The inclusion into account to maximize success rates
and exclusion criteria were meticulously defined to (ii) Real-time decision support: AI offers clinicians
ensure a thorough search. Only publications that were timely insights during critical procedures, such as
written in the English language were considered. The ET, enhancing accuracy and efficacy
studies spanned from June 1, 2015, to January 9, 2024, (iii) Predictive analytics: AI accurately predicts
considering that older studies might not reflect recent treatment outcomes, enabling proactive adjustments
technological advancements. to optimize success rates
(iv) Automated laboratory processes: AI automates
4.4.1. Factors influencing analytical results tasks such as sperm and embryo analysis, enhancing
One of the most concerning limitations of this reviewed efficiency and reducing errors
study is sampling imbalance. Some papers in this study do (v) Remote monitoring and teleconsultation:
not adequately represent the diverse population undergoing AI-powered systems enable continuous patient
ART, creating potential bias. Some AI models were trained monitoring and teleconsultation, extending ART
using datasets of a limited number of clinics, thereby accessibility
lacking generalizability. Another limitation is the variation (vi) Genomic analysis: AI identifies genetic risks, aiding
of validation techniques used across studies. Differences informed decisions on embryo selection and genetic
in validation techniques can influence performance screening for better outcomes
matrices. Without external validation, these models may (vii) Enhanced quality control: AI ensures optimal
demonstrate high accuracy on training data but perform laboratory conditions, minimizing variability and
poorly in real-world applications. Furthermore, variability enhancing success rates.
in model predictions, influenced by various characteristics The integration of AI into these areas promises
of the dataset provided, raises concerns about clinical personalized, efficient, and accessible reproductive
reliability. AI models are often trained on retrospective healthcare solutions, revolutionizing the field of
data (data collected from past events or historical records), reproductive medicine. Furthermore, robust regulatory
and their performance in real-world clinical settings raises frameworks are essential to guarantee the ethical and safe
concerns. In addition, some AI models lack interpretability application of AI in ART, highlighting the requirement for
(black-box algorithms). This further complicates the policies and supervision to promote ethical AI adoption in
integration of AI models into ART. reproductive medicine. Moreover, the utilization of AI in
89

Volume 2 Issue 3 (2025) 16 https://doi.org/10.36922/aih.7170