source, and batteries. A smartphone application was developed which performed the analysis locally. The AI utilized by the application was transfer-learned, trained, and validated with 1790 embryo images. To test our system, 50 embryos donated by patients were imaged using the smartphone system at 113 hpi. Images were automatically analyzed by our developed network without the need for any image processing. Performance metrics were calculated for the smartphone system and the overall performance of the smartphone system with the performance of deep-learning based approach that used Embryoscope data was compared.RESULTS: The accuracy of such a system in classifying 50 embryos based on their blastocyst status was 96% (CI: 86.29% to 99.51%). Its sensitivity and specificity were 93.55% (CI: 78.58% to 99.21%) and 100% (CI: 82.35% to 100%), respectively, while its positive and negative predictive values were 100% and 90.48% (CI: 71.32% to 97.32%), respectively. A chi-squared analysis comparing the performance of an Embryoscope-based deep-learning approach with our smartphone system-based deep-learning approach revealed an insignificant difference of 5.03% (P¼0.33, P>0.05).CONCLUSIONS: The results reported here demonstrate that combined with the use of an AI-empowered imaging system, automated embryo analysis is not limited to only expensive time-lapse hardware and inexpensive (<$100) systems can be developed for use at fertility centers without loss in performance. The overall impact of our AI-empowered system is significant since it enables integration into clinical practices at resource-limited settings at very minimal costs.
OBJECTIVE: The field of IVF has focused on embryo selection technology and multiple methods have been utilized including metabolomics, time lapse imaging and PGT-A. PGT-A was touted as the optimal method but it has recently come under scrutiny due to some evidence of euploid births from embryos found to be aneuploid on testing [1][2][3]. A selection method can only enhance the chances of success if we have a cohort of embryos to select from and yet, we have inadequate counseling tools for patients on their chances of having supernumerary embryos for a selection method to be applicable. We have identified factors predictive of having supernumerary embryos in freeze-all cycles. Therefore, we sought to create a clinical prediction model using those identified factors for clinical counseling.DESIGN: Retrospective cohort study of women who underwent freeze-all cycles in 2014.MATERIALS AND METHODS: Data were obtained from the Society for Assisted Reproductive Technology. We defined supernumerary as having two or more embryos cryopreserved. We utilized previously identified predictors and entered them into a logistic regression model presenting a receiver operating characteristic curve (ROC) for all predictors. Any predictor that did not alter the area under the curve for the ROC was removed from the prediction model. We then utilized methods described by Sullivan and colleagues [4] to modify the final model into a risk index. The number of points assigned to each significant covariate equaled its regression coefficient divided by the parameter estimated in the model with the smallest value rounded to the nearest whole number. The accuracy of the prediction model was then tested using an ROC.RESULTS: Of 31,537 freeze-all cycles in 2014, 18,250 produced supernumerary embryos. We included 16,395 cycles into the logistic regression model after excluding cycles with missing AMH as this was a very strong predictor of the outcome. Table 1 demonstrates the points assigned to each necessary predictor. The area under the curve (AUC) for the ROC was 0.84.CONCLUSIONS: Age, AMH and number of eggs retrieved are necessary predictors for the model. The AUC for the ROC is considered excellent discrimination and therefore, this model can be used to counsel patients undergoing freeze-all cycles on their probability of having supernumerary embryos for a selection method to be applicable.
No abstract
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.