Page 10 - AIH-2-3

Artificial Intelligence in Health AI in embryo selection for ART

(iv) Articles that do not meaningfully discuss AI 3.1. Data overview
application within IVF procedures Between June 1, 2015, and January 9, 2024, two review
(v) Studies conducted on laboratory animals writers (ASR and ABR) thoroughly searched across
(vi) Non-peer-reviewed sources, such as editorials, four databases: PubMed, Scopus, Google Scholar, and
opinions, and non-scholarly articles IEEE Xplore. A date restriction was applied to exclude
(vii) Studies not published in English; this is to ensure outdated models from the early stages of AI development,
accessibility for analysis.
ensuring the relevance of the technologies to the current
2.6. Search strategy AI landscape. The initial search yielded 656 articles from
Scopus, 249 from PubMed, four from IEEE Xplore, and
The search strings presented in Table 1 were used to five from Google Scholar. After removing 85 duplicates
identify all relevant articles and documents. Initially, the using Microsoft Excel (Microsoft, United States of America
first search string was applied, yielding 56 results from [USA]), a total of 829 articles remained. Following a title
Scopus, 98 from PubMed, and 3 from the IEEE Xplore and abstract screening, 789 articles were excluded, leaving
database. Then, the search string was modified to achieve 40 articles for eligibility assessment. Ten articles were
better results. subsequently excluded due to issues with data extraction,
2.7. Study selection process non-English language, lack of linkage with AI, or poor
technical implementation. The study selection process
First, research questions were developed, followed by a is illustrated in Figure 2. Ultimately, 30 articles that met
search string. Three researchers (ABR, ASR, and AMS) the inclusion criteria were retained for data extraction, as
performed the initial database search and removed summarized in Table 2.
duplicate entries. Two researchers (ASR and ABR)
reviewed all collected abstracts using the inclusion and 3.2. Risk of bias (RoB) assessment
exclusion criteria. Senior researcher AMS assessed articles The systematic review assessed the RoB in the included
with disagreements to establish consensus on decisions. studies to evaluate the validity and reliability of the results.

2.8. Study selection and bias control RoB was evaluated across multiple domains, such as
selection bias, performance bias, detection bias, attrition
The selection approach utilized a combination of bias, and reporting bias, using well-developed tools, such
engineering and health science datasets to enhance as the Cochrane RoB2 Tool for randomized controlled
reliability and minimize publication bias. Two researchers trials and the ROBINS-I tool for non-randomized studies.
(ABR and ASR) reviewed the titles and abstracts to reduce Most studies relied on retrospective data, which has the
selection bias, while senior researcher AMS meticulously potential for bias due to the non-randomized selection
analyzed a paper to identify errors and further mitigate bias. of participants. For example, studies such as those by
16
17
3. Results Theilgaard Lassen et al. and Cimadomo et al. employed
internal validation methods, which limit generalizability.
This section outlines the key findings and emerging Meanwhile, several studies, such as those by Johansen
patterns identified through the systematic review. It et al. and Bori et al., implemented AI models trained
18
19
discusses the implications of these results by comparing on time-lapse imaging data without clearly defined
them with previous research and highlighting recent standard protocols. The absence of standardized protocols
developments in the field. Furthermore, the review across clinics could have introduced heterogeneity in
examines any limitations encountered during the process data collection and analysis, thereby affecting the results.
and considers their potential influence on the outcomes. In addition, in AI-based embryo selection, as seen in

Table 1. Keywords and search items

S. No. Keywords and search items Number of publications from database
Scopus PubMed IEEE Xplore
1. ((AI) AND (ART) AND (IVF)) 56 98 3
2. (((AI) OR (artificial intelligence) OR (machine learning) OR (deep learning)) AND 656 249 4
((embryo selection) OR (blastocyst transfer) OR (preimplantation genetic diagnosis)) AND
((assisted reproductive technologies) OR (in vitro fertilization) OR (Intracytoplasmic Sperm
Injection) OR (Gamete Intrafallopian Transfer) OR (Zygote Intrafallopian Transfer))) AND
((precision medicine) OR (predictive algorithms) OR (prognosis)))

Volume 2 Issue 3 (2025) 4 https://doi.org/10.36922/aih.7170