AI Analysis Reveals Optimal Follicle Sizes for Improved IVF Success Rates

AI could improve the success of IVF treatment

During IVF treatment, doctors use ultrasound scans to monitor the size of follicles -- small sacs in the ovaries containing eggs -- to decide when to give a hormone injection known as the 'trigger' to prepare the eggs for collection and ensure that they are ready to be fertilised with sperm to create embryos. The timing of the trigger is a key decision, as it works less effectively if the follicles are too small or too large at the time of administration. After the eggs are collected and fertilised by sperm, an embryo is then selected and implanted into the womb to hopefully lead to pregnancy. Researchers used 'Explainable AI' techniques -- a type of AI that allows humans to understand how it works -- to analyse retrospective data on more than 19,000 patients who had completed IVF treatment. They explored which follicle sizes were associated with improved rates of retrieving mature eggs to result in babies being born. They found that delivering the hormone injection when a greater proportion of follicles were sized between 13-18mm was linked to higher rates of mature eggs being retrieved and improved rates of babies being born. Currently, clinicians use ultrasound scans to measure the size of the lead (largest) follicles and generally give the trigger injection when a threshold of either two or three lead follicles greater than 17 or 18mm has been reached. Their findings suggest that maximising the proportion of intermediately sized follicles could optimise the number of mature eggs retrieved and improve the rates of babies being born. The team believe that the findings from the study highlight the potential of AI to aid in the personalisation of IVF treatment to improve clinical outcomes for patients and maximise their chance of taking home a baby. The team plan to create an AI tool that will utilise findings from their research to personalise IVF treatment and support clinicians' decision making at each step of the IVF process. They will apply for funding to take this tool into clinical trials. The research, published in Nature Communications, is led by researchers at Imperial College London, University of Glasgow, University of St Andrews, and clinicians at Imperial College Healthcare NHS Trust. It is funded by UK Research and Innovation and the National Institute for Health and Care Research (NIHR) Imperial Biomedical Research Centre (BRC). Dr Ali Abbara, NIHR Clinician Scientist at Imperial College London and Consultant in Reproductive Endocrinology at Imperial College Healthcare NHS Trust, and co-senior author of the study said: "IVF provides help and hope for many patients who are unable to conceive but it's an invasive, expensive, and time-consuming treatment. It can be heartbreaking when it fails, so it's important to ensure that this treatment is as effective as possible. "AI can offer a new paradigm in how we deliver IVF treatment and could lead to better outcomes for patients. "IVF produces so much rich data that it can be challenging for doctors to fully make use of all of it when making treatment decisions for their patients. Our study has shown that AI methods are well suited to analysing complex IVF data. In future, AI could be used to provide accurate recommendations to improve decision-making and aid in personalisation of treatment, so that we can give each couple the very best possible chance of having a baby." Professor Waljit Dhillo, an NIHR Senior Investigator from the Department of Metabolism, Digestion and Reproduction at Imperial College London, Consultant Endocrinologist at Imperial College Healthcare NHS Trust and co-senior author of the study, added: "Our findings could pave the way for a new approach to maximise the success of IVF treatment, leading to more pregnancies and births. "Our study is the first to analyse a large dataset to show that AI can identify the specific follicle sizes that are most likely to yield mature eggs more precisely than current methods. "This is an exciting development as the findings suggest that we can use information from a much wider set of follicle sizes to decide when to give patients trigger shots rather than just the size of only the largest follicles -- which is what is used in current clinical practice." Dr Thomas Heinis, co-senior author from the Department of Computing at Imperial College London, added: "Explainable AI can be a valuable resource in healthcare. Where the stakes are so high for making the best possible decision, this technique can support doctors' decision making and lead to better outcomes for patients. Importantly, we expect computing power to improve exponentially in the near future, enabling us to make decisions using precise data in a way that hasn't been possible previously." One in six couples experience infertility and IVF has emerged as a valuable intervention to help patients conceive. Trigger injection A key decision in IVF treatment is when to use the 'trigger' shot of hormones, such as human chorionic gonadotropin (hCG), to mature eggs for collection. The timing of the trigger shot impacts on the number of mature eggs retrieved and the success of treatment. Clinicians use ultrasound scans to measure the size of the lead (largest) follicles. They will generally give the trigger shot when a threshold of either two or three lead follicles greater than 17 or 18mm in diameter has been reached. However, this method lacks precision and does not take into consideration the sizes of each individual follicle and their likelihood of each follicle yielding a mature egg. Follicle sizes In the retrospective study, the team used AI techniques on data from 19,082 patients aged between 18-49 years of age who had treatment in one of 11 clinics across the UK- including IVF clinics at Imperial College Healthcare NHS Trust -- and two in Poland between 2005-2023. They examined individual follicle sizes on the days prior to and on the day of trigger administration. The researchers found that intermediately sized follicles of 13-18mm were associated with more mature eggs subsequently being retrieved. The data suggested that having a greater number of follicles within this range on the day of trigger was associated with better clinical outcomes. They also found that stimulating the ovaries for too long, such that there was a greater number of larger follicles (more than 18mm) on the day of trigger administration, could lead to a premature elevation of the hormone progesterone. This can have a negative impact on IVF outcomes by affecting the proper development of the endometrium -- the tissue that line the uterus and is important for an embryo implanting to result in pregnancy. This reduces the chances of an embryo implanting and subsequently leading to a live birth. These AI derived insights can help the team develop evidence-based IVF protocols guided by data that should improve the efficiency of treatment.

AI could be used to improve chances of pregnancy through IVF

Researchers looked at how artificial intelligence could be used to improve the outcome of an IVF procedure. Artificial intelligence (AI) could be used to help boost the success rate of in vitro fertilisation (IVF), according to a new study. IVF involves removing an egg from a woman's ovaries and fertilising it in a laboratory with sperm. If the fertilisation is successful, the embryo can then be placed in a uterus, where babies develop. The eggs, contained in small sacs called follicles, are prepared with an injection of a hormone called the "trigger" ahead of collection. Using data from over 19,000 patients and AI, researchers discovered that delivering injection when a greater proportion of follicles were sized between 13 and 18 mm was linked to improved rates of getting mature eggs and future births. Currently, ultrasound imaging is used to assess the dimensions of the largest follicles. "IVF produces so much rich data that it can be challenging for doctors to fully make use of all of it when making treatment decisions for their patients. Our study has shown that AI methods are well suited to analysing complex IVF data," Dr Ali Abbara, a reader in endocrinology at Imperial College London and co-senior author of the study, said in a statement.** "In future, AI could be used to provide accurate recommendations to improve decision-making and aid in personalisation of treatment, so that we can give each couple the very best possible chance of having a baby," he added. The team published their findings in the journal Nature Communications and is planning to leverage this research to create an AI tool and attempt to take it into clinical trials. "Explainable AI can be a valuable resource in healthcare. Where the stakes are so high for making the best possible decision, this technique can support doctors' decision-making and lead to better outcomes for patients," said Dr Thomas Heinis, a reader in computing at Imperial College London and one of the study's authors. "Importantly, we expect computing power to improve exponentially in the near future, enabling us to make decisions using precise data in a way that hasn't been possible previously". Infertility affects one in six couples, according to the World Health Organization (WHO). The average success rate of the treatment, meaning that it results in a live birth, ranged from 32 per cent for women under 35 to 4 per cent for women over 44 in 2019, according to the UK National Health Service (NHS). Infertility doesn't only concern women, with studies showing an increase in male infertility since the mid-1970s. One in 20 men currently faces reduced fertility with researchers saying that potential causes include exposure to certain chemicals, "rising rates of obesity, and the trend of delayed parenthood".

Explainable AI techniques reveal ideal follicle sizes for successful IVF treatments

During IVF treatment, doctors use ultrasound scans to monitor the size of follicles -- small sacs in the ovaries containing eggs -- to decide when to give a hormone injection known as the "trigger" to prepare the eggs for collection and ensure that they are ready to be fertilized with sperm to create embryos. The timing of the trigger is a key decision, as it works less effectively if the follicles are too small or too large at the time of administration. After the eggs are collected and fertilized by sperm, an embryo is then selected and implanted into the womb to hopefully lead to pregnancy. Researchers used "Explainable AI" techniques -- a type of AI that allows humans to understand how it works -- to analyze retrospective data on more than 19,000 patients who had completed IVF treatment. They explored which follicle sizes were associated with improved rates of retrieving mature eggs to result in babies being born. They found that delivering the hormone injection when a greater proportion of follicles were sized between 13-18mm was linked to higher rates of mature eggs being retrieved and improved rates of babies being born. Currently, clinicians use ultrasound scans to measure the size of the lead (largest) follicles and generally give the trigger injection when a threshold of either two or three lead follicles greater than 17 or 18mm has been reached. Their findings suggest that maximizing the proportion of intermediately-sized follicles could optimize the number of mature eggs retrieved and improve the rates of babies being born. The team believe that the findings from the study highlight the potential of AI to aid in the personalization of IVF treatment to improve clinical outcomes for patients and maximize their chance of taking home a baby. The team plan to create an AI tool that will utilize findings from their research to personalize IVF treatment and support clinicians' decision making at each step of the IVF process. They will apply for funding to take this tool into clinical trials. The research, published in Nature Communications, is led by researchers at Imperial College London, University of Glasgow, University of St Andrews, and clinicians at Imperial College Healthcare NHS Trust. Dr. Ali Abbara, NIHR Clinician Scientist at Imperial College London and Consultant in Reproductive Endocrinology at Imperial College Healthcare NHS Trust, and co-senior author of the study said, "IVF provides help and hope for many patients who are unable to conceive, but it's an invasive, expensive, and time-consuming treatment. It can be heartbreaking when it fails, so it's important to ensure that this treatment is as effective as possible. "AI can offer a new paradigm in how we deliver IVF treatment and could lead to better outcomes for patients. "IVF produces so much rich data that it can be challenging for doctors to fully make use of all of it when making treatment decisions for their patients. Our study has shown that AI methods are well suited to analyzing complex IVF data. "In the future, AI could be used to provide accurate recommendations to improve decision-making and aid in personalization of treatment, so that we can give each couple the very best possible chance of having a baby." Professor Waljit Dhillo, an NIHR Senior Investigator from the Department of Metabolism, Digestion and Reproduction at Imperial College London, Consultant Endocrinologist at Imperial College Healthcare NHS Trust and co-senior author of the study, added, "Our findings could pave the way for a new approach to maximize the success of IVF treatment, leading to more pregnancies and births. "Our study is the first to analyze a large dataset to show that AI can identify the specific follicle sizes that are most likely to yield mature eggs more precisely than current methods. "This is an exciting development as the findings suggest that we can use information from a much wider set of follicle sizes to decide when to give patients trigger shots rather than just the size of only the largest follicles -- which is what is used in current clinical practice." Dr. Thomas Heinis, co-senior author from the Department of Computing at Imperial College London, added, "Explainable AI can be a valuable resource in health care. Where the stakes are so high for making the best possible decision, this technique can support doctors' decision-making and lead to better outcomes for patients. Importantly, we expect computing power to improve exponentially in the near future, enabling us to make decisions using precise data in a way that hasn't been possible previously." One in six couples experience infertility and IVF has emerged as a valuable intervention to help patients conceive. Trigger injection A key decision in IVF treatment is when to use the "trigger" shot of hormones, such as human chorionic gonadotropin (hCG), to mature eggs for collection. The timing of the trigger shot impacts on the number of mature eggs retrieved and the success of treatment. Clinicians use ultrasound scans to measure the size of the lead (largest) follicles. They will generally give the trigger shot when a threshold of either two or three lead follicles greater than 17 or 18mm in diameter has been reached. However, this method lacks precision and does not take into consideration the size of each individual follicle and the likelihood of each follicle yielding a mature egg. Follicle sizes In the retrospective study, the team used AI techniques on data from 19,082 patients aged between 18-49 years of age who had treatment in one of 11 clinics across the UK -- including IVF clinics at Imperial College Healthcare NHS Trust -- and two in Poland between 2005-2023. They examined individual follicle sizes on the days prior to and on the day of trigger administration. The researchers found that intermediately-sized follicles of 13-18mm were associated with more mature eggs subsequently being retrieved. The data suggested that having a greater number of follicles within this range on the day of trigger was associated with better clinical outcomes. They also found that stimulating the ovaries for too long, such that there was a greater number of larger follicles (more than 18mm) on the day of trigger administration, could lead to a premature elevation of the hormone progesterone. This can have a negative impact on IVF outcomes by affecting the proper development of the endometrium -- the tissue that lines the uterus and is important for an embryo implanting to result in pregnancy. This reduces the chances of an embryo implanting and subsequently leading to a live birth. These AI-derived insights can help the team develop evidence-based IVF protocols guided by data that should improve the efficiency of treatment.

AI analysis reveals optimal follicle sizes for IVF success

Imperial College LondonJan 8 2025 During IVF treatment, doctors use ultrasound scans to monitor the size of follicles - small sacs in the ovaries containing eggs - to decide when to give a hormone injection known as the 'trigger' to prepare the eggs for collection and ensure that they are ready to be fertilized with sperm to create embryos. The timing of the trigger is a key decision, as it works less effectively if the follicles are too small or too large at the time of administration. After the eggs are collected and fertilized by sperm, an embryo is then selected and implanted into the womb to hopefully lead to pregnancy. Researchers used 'Explainable AI' techniques - a type of AI that allows humans to understand how it works - to analyze retrospective data on more than 19,000 patients who had completed IVF treatment. They explored which follicle sizes were associated with improved rates of retrieving mature eggs to result in babies being born. They found that delivering the hormone injection when a greater proportion of follicles were sized between 13-18mm was linked to higher rates of mature eggs being retrieved and improved rates of babies being born. Currently, clinicians use ultrasound scans to measure the size of the lead (largest) follicles and generally give the trigger injection when a threshold of either two or three lead follicles greater than 17 or 18mm has been reached. Their findings suggest that maximizing the proportion of intermediately sized follicles could optimize the number of mature eggs retrieved and improve the rates of babies being born. The team believe that the findings from the study highlight the potential of AI to aid in the personalization of IVF treatment to improve clinical outcomes for patients and maximize their chance of taking home a baby. The team plan to create an AI tool that will utilize findings from their research to personalize IVF treatment and support clinicians' decision-making at each step of the IVF process. They will apply for funding to take this tool into clinical trials. The research, published in Nature Communications, is led by researchers at Imperial College London, University of Glasgow, University of St Andrews, and clinicians at Imperial College Healthcare NHS Trust. It is funded by UK Research and Innovation and the National Institute for Health and Care Research (NIHR) Imperial Biomedical Research Centre (BRC). Dr Ali Abbara, NIHR Clinician Scientist at Imperial College London and Consultant in Reproductive Endocrinology at Imperial College Healthcare NHS Trust, and co-senior author of the study said: "IVF provides help and hope for many patients who are unable to conceive but it's an invasive, expensive, and time-consuming treatment. It can be heartbreaking when it fails, so it's important to ensure that this treatment is as effective as possible. "AI can offer a new paradigm in how we deliver IVF treatment and could lead to better outcomes for patients. "IVF produces so much rich data that it can be challenging for doctors to fully make use of all of it when making treatment decisions for their patients. Our study has shown that AI methods are well suited to analysing complex IVF data. In future, AI could be used to provide accurate recommendations to improve decision-making and aid in personalisation of treatment, so that we can give each couple the very best possible chance of having a baby." Professor Waljit Dhillo, an NIHR Senior Investigator from the Department of Metabolism, Digestion and Reproduction at Imperial College London, Consultant Endocrinologist at Imperial College Healthcare NHS Trust and co-senior author of the study, added: "Our findings could pave the way for a new approach to maximise the success of IVF treatment, leading to more pregnancies and births. "Our study is the first to analyse a large dataset to show that AI can identify the specific follicle sizes that are most likely to yield mature eggs more precisely than current methods. "This is an exciting development as the findings suggest that we can use information from a much wider set of follicle sizes to decide when to give patients trigger shots rather than just the size of only the largest follicles - which is what is used in current clinical practice." Explainable AI can be a valuable resource in healthcare. Where the stakes are so high for making the best possible decision, this technique can support doctors' decision making and lead to better outcomes for patients. Importantly, we expect computing power to improve exponentially in the near future, enabling us to make decisions using precise data in a way that hasn't been possible previously." Dr. Thomas Heinis, co-senior author, Department of Computing, Imperial College London One in six couples experience infertility and IVF has emerged as a valuable intervention to help patients conceive. Trigger injection A key decision in IVF treatment is when to use the 'trigger' shot of hormones, such as human chorionic gonadotropin (hCG), to mature eggs for collection. The timing of the trigger shot impacts on the number of mature eggs retrieved and the success of treatment. Clinicians use ultrasound scans to measure the size of the lead (largest) follicles. They will generally give the trigger shot when a threshold of either two or three lead follicles greater than 17 or 18mm in diameter has been reached. However, this method lacks precision and does not take into consideration the sizes of each individual follicle and their likelihood of each follicle yielding a mature egg. Follicle sizes In the retrospective study, the team used AI techniques on data from 19,082 patients aged between 18-49 years of age who had treatment in one of 11 clinics across the UK- including IVF clinics at Imperial College Healthcare NHS Trust - and two in Poland between 2005-2023. They examined individual follicle sizes on the days prior to and on the day of trigger administration. The researchers found that intermediately sized follicles of 13-18mm were associated with more mature eggs subsequently being retrieved. The data suggested that having a greater number of follicles within this range on the day of trigger was associated with better clinical outcomes. They also found that stimulating the ovaries for too long, such that there was a greater number of larger follicles (more than 18mm) on the day of trigger administration, could lead to a premature elevation of the hormone progesterone. This can have a negative impact on IVF outcomes by affecting the proper development of the endometrium - the tissue that line the uterus and is important for an embryo implanting to result in pregnancy. This reduces the chances of an embryo implanting and subsequently leading to a live birth. These AI derived insights can help the team develop evidence-based IVF protocols guided by data that should improve the efficiency of treatment. Imperial College London Journal reference: Hanassab, S., et al. (2025). Explainable artificial intelligence to identify follicles that optimize clinical outcomes during assisted conception. Nature Communications. doi.org/10.1038/s41467-024-55301-y.

Explainable artificial intelligence to identify follicles that optimize clinical outcomes during assisted conception - Nature Communications

Follicles sized 13-18 mm contributed most to the number of MII oocytes retrieved and therefore we analyzed whether the proportion of follicles within this size range was associated with the live birth rate (LBR) after fresh embryo transfer. LBR was 30.48% (95% CI: 29.68%-31.29%) in the patient cohort (n = 12,724). Using a logistic regression model (n = 9209), we found that the proportion of follicles within 13-18 mm on the DoT was positively associated with LBR (OR: 1.03 (1.00-1.06) per 10% points change; p = 0.048) when adjusted for age, total follicle count, and type of trigger administered (Fig. 5a). We next examined whether the mean follicle size impacted on LBR (Fig. 5b), and found a negative association (OR: 0.95 (0.93-0.98) per 1 mm change; p = 0.001). We found that LBR (n = 427) was reduced with progesterone elevation on the DoT (Fig. 5c), whilst mature oocyte yield (n = 646) remained similar. Further, serum progesterone on the DoT (n = 994) was increasingly elevated as the number of follicles sized >18 mm on the DoT was greater (Fig. 5d). We present findings from a large multi-center European study incorporating data from 19,082 patients utilizing XAI techniques to establish the relationship between follicle sizes at the end of ovarian stimulation (OS) and the subsequent retrieval of mature oocytes, as well as relevant downstream clinical outcomes. We showed that a contiguous range of follicles sized 13-18 mm contributed most to the number of mature oocytes subsequently retrieved. In comparison to using lead follicle size to inform trigger timing, our data suggests that maximizing the proportion of follicles within this size range could further optimize the number of mature oocytes retrieved to improve IVF outcomes, including LBR, pending prospective evaluation. Furthermore, extending the duration of OS resulted in a greater number of larger follicles (>18 mm) on the DoT that not only contributed less to the yield of mature oocytes but also resulted in premature progesterone elevation with a consequent negative impact on live birth, likely due to its adverse effect on the endometrial stage. These data highlight the potential of ML methods to aid in personalizing ART treatment to optimize clinical outcomes. Our results are consistent with a previous pilot study using a much smaller sample size of 499 patients that found that follicles sized 12-19 mm on DoT were most contributory to the number of mature oocytes. Hariton et al. found that an input feature containing the number of follicles sized 16-20 mm on the DoT was most contributory to the model performance using an ensemble ML model of similar complexity in data from a single clinic comprising 7866 ICSI treatment cycles with various protocol types. Similarly, Reuvenny et al. used an XGBoost model on data from GnRH antagonist ("short" protocol) co-treated cycles from a single center (n = 3599 treatment cycles) to show that an input variable of 14-16 mm sized follicles on the DoT highly contributed towards model performance. Further, Fanton et al. developed a linear regression model with data from three clinics in the USA (n = 30,278 treatment cycles) that identified an input variable of 14-15 mm sized follicles on the DoT as the most important, followed by 16-17 mm follicle sizes. In our study, we examined individual follicle sizes to identify the group of follicles that contribute most to the retrieval of mature oocytes and, in turn, downstream clinical outcomes. To validate the methodology further, we also analyzed follicle sizes that were most important on the days prior to DoT. We showed that follicles sized 10-15 mm on DoT-2 and 12-17 mm on DoT-1 were the most contributory to predicting MII oocyte yield on the days preceding DoT. This smooth trajectory of optimal follicle size range shift corroborated well with expectations in mean follicular growth rates per day during OS. The aforementioned ML studies also showed agreement, where the most important input variables to the models utilizing follicle sizes dropped by 1-2 mm in range a day prior to the DoT. In order to examine whether similar follicle sizes were important in predicted good ovarian response patients, we stratified our data by age. We found that follicles sized 13-18 mm were most contributory in younger patients aged ≤35 years, but those 11-20 mm (particularly 15-18 mm) were most important in patients >35 years. In line with our data whereby increased mean follicle size by the end of OS was associated with a negative impact on LBR, it has been suggested that older patients with diminished ovarian reserve may benefit from earlier trigger administration in modified natural cycles. To date, the impact of age on trigger timing in IVF cycles remains uncertain. Likewise, although most patients are required to be non-smokers to access state-funded care, there was insufficient recording of smoking status to formally assess any impact of smoking. It has been hypothesized that due to the preceding period of gonadotropin suppression that could synchronize follicle growth, the GnRH agonist co-treated ("short") protocol could lead to a more uniform cohort of follicles than in the GnRH antagonist co-treated ("long") protocol. We found that larger follicles sized 14-20 mm were more contributory to the yield of mature oocytes in GnRH agonist co-treated protocol cycles whereas follicles sized 12-19 mm were most important in GnRH antagonist protocol cycles. A meta-analysis of seven randomized control trials (n = 1295 patients) comparing protocol types found that patients undergoing the "long" protocol had a significantly higher number of oocytes retrieved (p < 0.00001) in those that received a delayed hCG trigger by 24-48 h after OS. Our data are consistent with these previous trials, suggesting that larger follicles may be associated with improved oocyte yield in the GnRH agonist co-treated protocol. Several points of strength should be highlighted in our presented study. Firstly, we used XAI techniques in our study. Explainability is currently a paramount characteristic of AI decision-making in ART clinics and the introduction of data-driven insights that align with clinical reasoning promotes the possible wider adoption of clinical decision-support systems in the ART domain. Secondly, the use of data across two countries comprising eleven clinics presents a heterogeneous patient population with a variety of clinical practices and treatment protocols. Since many clinics are involved, the use of internal-external validation at this stage of development was appropriate, whereby the developed models and their performance metrics are provided with a standard deviation due to each clinic behaving as an independent test set to observe permutation importance and model error. Furthermore, we chose to avoid random splits of data from the same clinics in both training and test sets when reporting outcome metrics, as this has been shown to result in bias and unknown generalizability ("transportability") to wider patient populations. Thirdly, by only using individual follicle sizes as separate input variables (i.e., not in a grouped/binned fashion), a contiguous range of follicles that contribute most towards mature oocyte yield was more explicitly identified. Finally, we ensured only to include the first treatment cycle of each patient in this study for model development and validation (i.e., n = 19,082 represents both the number of treatment-naive patients and cycles). This was to avoid auto-correlation between successive cycles of a single patient (e.g., a clinician's decision-making is likely to be influenced by a previous treatment cycle) since longitudinal random effects across sequential cycles can influence the permutation importance. Similarly, input parameters that present multi-collinearity (e.g., estradiol and follicle size) can also result in unreliable insights from permutation importance analysis. Variability in follicle size measurements, both within and between observers, has been a documented challenge in ART. We observed a few cases where the total number of oocytes retrieved exceeded the follicles recorded at the time of trigger. This discrepancy underscores the likelihood that smaller follicles are often not consistently recorded, a practice echoed in anecdotal discussions with clinicians, who frequently attribute this to the lower probability of such follicles yielding oocytes. A previous study has shown that follicle imputation has limited impact on enhancing model performance in this specific context. Although excluding such treatment cycles could artificially improve model error, it is then not possible to build models that are more robust to such measurement error in future applications. Our approach utilizes an ensemble-based ML model with Bayesian optimization and can mitigate against the impact of data inconsistencies due to extremes of biological variation and/or measurement error. By employing a loss metric like mean absolute error (MAE), which is less sensitive to outliers, the model offers a more robust analysis validated across multiple clinics. The need for more objective ultrasound scanning methods, potentially through the integration of automated algorithms, may further improve the accuracy and reliability of follicle measurements and associated algorithms in ART. Our data suggest that a novel approach to deciding when to administer the trigger of oocyte maturation could be based on the proportion of intermediately-sized follicles (e.g., 13-18 mm) rather than the traditional threshold-based approach assessing when 2-3 lead follicles reach 17 or 18 mm in size (Fig. 4). Further, our data also suggests that mean follicle size could impact LBR in fresh embryo transfer cycles (Fig. 5), aside from a direct effect on the ability to yield oocytes, potentially via larger follicles resulting in premature progesterone elevation. A prospective randomized controlled trial is required to demonstrate the benefit of our new proposed approach to determine trigger administration based on the entire follicle cohort in comparison to the current threshold-based approach. Since our data demonstrates a possible trade-off between LBR and delayed trigger administration, a trial comparing trigger strategies on this basis would therefore be potentially more definitive in determining the impact of trigger timing on clinical outcomes. It should be noted that although a range of follicle sizes that contribute relatively more than others varied marginally depending on the patient stratifications considered (Fig. 2). Ultimately, an ML model that considers individual follicle sizes and their relative contributions, in addition to patient characteristics, could be harnessed as part of a clinical decision support system. In conclusion, we establish that intermediately-sized follicles on the day of trigger contribute the most to the retrieval of mature oocytes and subsequent embryo development. Utilizing the sizes of all follicles, rather than just the size of only the lead follicles, could offer a target for OS protocols and inform the timing of trigger administration to optimize clinical outcomes. These data highlight the potential of XAI techniques to provide data-driven optimization of IVF treatment to improve clinical outcomes.