Exploring Traditional Machine Learning and Deep Learning for Predicting Intracytoplasmic Sperm Injection (ICSI) Success Rates

The utilization of Intra-cytoplasmic sperm injection (ICSI) in treating infertility amongcouples has significantly advanced patient care [۱]. Despite its considerable expense, the success ratesof ICSI remain disappointingly low, leading to significant anxiety for couples facing failed pregnancies.One of the effective ways to address this problem is to use state-of-the art machine learning algorithms[۲-۵]. This study endeavors to identify key predictors for forecasting the success of ICSI and enhanceprediction accuracy through the application of support vector machines (SVM) and deep learningtechniques.A cohort of ۳۴۵ patients undergoing ICSI treatment provided data comprising ۲۹ numerical and nominalfeatures. To address the issue of class imbalance within the dataset, Synthetic Minority OversamplingTechnique (SMOTE) was employed. Two distinct methodologies were applied to construct predictivemodels. The first approach involved utilizing a genetic algorithm (GA) for feature selection inconjunction with SVM for prediction. Employing ۳-fold cross-validation, the GA+SVM modelidentified ۴ significant features and achieved an average accuracy of ۰.۷۲۸. The second methodemployed a state-of-the-art deep neural network. To ensure the robustness of predictions, the modelunderwent training and testing iterations five times with varying random seeds. This approach yieldedan average accuracy of ۰.۷۳۷.The findings indicate that deep neural networks surpassed SVM in predictive performance,underscoring the superior efficacy of deep learning models in forecasting the success rates of ICSI. Thisstudy contributes to advancing the understanding of factors influencing ICSI outcomes and underscoresthe potential of deep learning techniques in optimizing fertility treatment strategies.

کلیدواژه ها:

Intra-cytoplasmic sperm injection (ICSI) ، Machine Learning (ML) ، Genetic Algorithm (GA) ، Deep Learning