Machine Learning-Based Prediction of First Trimester Down Syndrome Risk in East Asian Populations.

Area of Science:

• Genetics and Genomics
• Artificial Intelligence in Healthcare
• Maternal-Fetal Medicine

Background:

• Down syndrome is the most common chromosomal abnormality in newborns.
• It frequently results in developmental delays and congenital structural anomalies.
• Accurate early prediction is crucial for timely intervention and management.

Purpose of the Study:

• To identify an optimal machine learning model for predicting first-trimester Down syndrome risk.
• To enhance the accuracy and efficiency of Down syndrome screening in East Asian populations.
• To explore the predictive capabilities of various machine learning models and data balancing algorithms.

Main Methods:

• Utilized a dataset of 3,812 cases from Taipei Chang Gung Memorial Hospital (May 2018–Feb 2024).
• Input fourteen features, including maternal age and serum markers, into twelve machine learning models.
• Employed seven data-balancing algorithms and evaluated model performance using AUC, accuracy, precision, recall, and F1 scores.

Main Results:

• LightGBM with Random Undersampling (RUS) achieved the highest recall (0.84) for high-risk cases.
• Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) models combined with Random Oversampling (ROS) attained the highest Area Under the Curve (AUC) of 0.939.
• The ANN-ROS model demonstrated superior performance in classifying Down syndrome risk.

Conclusions:

• The proposed ANN model, utilizing deep neural networks and ROS, achieved an AUC of 0.939 and accuracy of 0.97.
• This model shows significant potential for accurate first-trimester Down syndrome risk prediction in East Asian populations.
• The findings suggest improved efficiency and accuracy in Down syndrome screening processes.