Iullia Diakova
ART-IVF Reproductive Health Clinic, RussiaTitle: Deep learning neural network model for predicting clinical pregnancy outcomes in in vitro fertilization protocols
Abstract
The increasing prominence of neural networks and AI models across diverse fields in medicine plays a pivotal role in extracting complex dependencies from data, making predictions, and aiding decision-making processes. In IVF, it is essential to conduct additional analyses and modeling of protocol data, especially embryological components, to enhance outcome predictions for the cycle and to develop personalized highly effective medical care.
Developing a Deep Learning Neural Network (DNN) Model: Integration of KPI data and KPI Score correlations into the DNN with 20 clinical and laboratory parameters for pregnancy prediction.
Model Training and Performance: Training on a comprehensive dataset from 2013-2023, with 3858 embryo transfer cases used for the model with implementation using the KERAS library for the model development.
Results of Validation and Specificity: Validation across 10 randomly shuffled data frames, demonstrating a 96.8% specificity and 63% accuracy in predicting clinical pregnancy. Occurrence of false-positive predictions - 15.5%. Average model accuracy of 67.9%, SD 5.4%, maximum accuracy of 83.1%. Further model refinement through additional training on a new dataset from 2018 to 2023 (1600 protocols), resulting in an average accuracy of 85.5% and a maximum cross-validated accuracy of 88.6%. Evaluation of model errors through the analysis of 168 protocols with PGT-A, revealing a 27% error rate in predicting clinical pregnancy. Superiority of the model's predictive accuracy (85.5%) compared to literature-reported expert predictions based on clinical and laboratory parameters (51%, range 43–59%). Model has accuracy and reproducibility comparable to state-of-the-art models in time-lapse imaging.
Conclusion: The developed model exhibits robust predictive capabilities in clinical pregnancy outcomes, surpassing traditional models of ML, offering potential applications in quality control as a tool for remote monitoring of IVF department performance and serving as a valuable quality control and decision support in IVF clinics.
Biography
Iullia Diakova is a Clinical Embryologist at ART-IVF Reproductive Health Clinic, Moscow. She has performed over 4000 IVF procedures with high success rates and have contributed to several reports on conferences and training courses of the Cooper Surgical corporation. She is an expert and travelling Embryologist to perform embryo biopsy for PGT in various IVF clinics across Russia and abroad. She is an Expert PGT specialist in First Genetics Lab, Moscow and Genetics and Reproductive Center “Genetico”, Moscow. She provides consultation and support to other embryologists on embryo biopsy and other IVF techniques as laboratory supervisor. She is an organizer of external audit to ensure testing and laboratory work consistency and compliance with regulatory standards.
Organization Location
1309 Coffeen Avenue, STE 1200, Sheridan, Wyoming 82801, USA
Mail us at
Program Enquiry
gynaecology@scientexconferences.com
General Enquiry
contact@scientexconferences.com
Registration query
finance@scientexconferences.com
gynaecology@scientexevents.org
Our Social Presence
Call us at
USA :(+1) 202 773 8420