Live Birth Prediction and Optimizing Protocols in IVF/ICSI Treatments: A Machine Learning Approach

Infertility is a common issue that affects many couples, making it difficult for them to conceive a child naturally. In Vitro Fertilization (IVF) is an assisted reproductive technology (ART) that offers hope for couples struggling to have a baby. Integrating AI and machine learning in ART represents a ground-breaking shift towards more precise, efficient, and patient-centered fertility treatments. It enhances the accuracy of current practices and opens new possibilities for advanced and personalized reproductive care, ultimately aiming to improve success rates and patient outcomes in fertility clinics around the world. The main objective of this thesis is to develop two distinct machine learning-based methods to predict the probability of a live birth. The first method, LETFC, relies on transfers and their associated features for each transfer. On the other hand, the second method, FCEQ, provides a comprehensive overview of all transfers based on the quality of embryos in the cycle. This method is unique and provides valuable insights, making this study more informative. The study includes implementing ISala Hospital's embryo scoring system, determining factors that impact successful pregnancy, analyzing independent variables, quantitatively analyzing oocytes and embryos, identifying the appropriate timing for ovum retrieval, and comparing the performance of three different machine learning algorithms in each method: Support Vector Machine (SVM), Random Forest (RF), and XGB (eXtreme Gradient Boosting). All three machine learning models performed well in both methods. Sometimes, one ML algorithm, such as Random Forest, performed better in specific evaluation metrics. One of the key outcomes of this research was implementing the scoring of Isala Hospital, which helped reduce the number of features while ensuring the reliability of the hospital's scoring system. Explainable AI tools were used to determine which variables had the most influence on the prediction and to provide insights into how they affect the model's conclusions. Our findings show that the quality of embryos, frozen embryo count, female age, and endometrial thickness have the highest influence on IVF outcomes in the LETFC method. The FCEQ Method predicts full cycle success by considering the quality of embryos, the female's age, and other relevant features. The most significant factors in this method are the weighted score, the average time between freezing and thawing embryos, the female's age, and the number of transferred cryo embryos. These insights provide a deeper understanding of how different variables influence outcomes and can guide clinicians to make more informed decisions during treatment.