An interpretable machine learning model for predicting the optimal day of trigger during ovarian stimulation

Artificial Intelligence for Ovarian Stimulation

“…Only one computer decision support system encompasses the whole stimulation cycle with day-to-day decisions, but exclusively for IVF controlled hyperstimulation [10]. The few other algorithm-based treatments ever published take into account only one of two specific steps of the stimulation cycle and quite exclusively for IVF purposes: the starting dose [7][8][9][10][11] or the criteria for ovulation triggering [12,13]. To date, StimXpert represents the only stimulation software available assuming both 1 -the entire treatment cycle, and 2 -all validated protocols, from mono-follicular classic stimulation to multi-follicular controlled hyperstimulation.…”

Quantitative approaches in clinical reproductive endocrinology

“…Even where ML techniques provide an ability to predict outcomes, some methodologies can remain unexplainable (‘black-box’) [ 26 ], such that mechanistic insights into the decision processes carried out by such models may not be evident. Others harness more interpretable methods e.g., random forests [ 29 , 30 ] or linear regression [ 31 ], where the most important predictors can be identified. For example, top predictors of live birth after IVF treatment included female partner age, anti-Müllerian hormone (AMH) [ 32 ], number of high-quality embryos, and serum estradiol level (reflective of cumulative follicle size and, in turn, the number of eggs that will be retrieved) on the day of administration of the trigger for oocyte maturation [ 33 ].…”

“…However, this approach assumes uniform growth of the follicles behind these lead follicles, rather than a more diverse set of follicle sizes [ 35 ]. By harnessing ML techniques such as bagged decision trees [ 58 ], random forests [ 30 ], and linear regression [ 31 ], found in the literature, the size of follicles on the day of trigger most likely to yield oocytes has been estimated, and indicates the potential to support the optimization of the timing of trigger administration during clinical workflows [ 39 ]. Identification of this follicle size range enables the quantification of oocyte maturation [ 29 ], and can provide a target for response to gonadotropins when evaluating response to ovarian stimulation.…”

Avocados, bananas, and pancakes: Does the oocyte maturity ratio predict live birth rates in in vitro fertilization/intracytoplasmic sperm injection?

“…However, this hypothesis is speculative because the study was not randomized and physicians managed patients according to their judgment. It remains to be seen whether future use of artificial intelligence in stimulation decisions such as trigger timing can improve oocyte maturity ratios and thereby increase live births (4). A high oocyte maturity ratio was associated with a statistically significantly lower live birth in the ICSI subsample than the reference group of average maturity ratio.…”