A Hybrid Artificial Intelligence Approach for Down Syndrome Risk Prediction in First Trimester Screening.

Area of Science:

• Computational biology and bioinformatics
• Medical artificial intelligence
• Genetics and genomics

Background:

• First trimester prenatal screening is crucial for detecting Down Syndrome (DS).
• Current methods for DS risk prediction require enhancement in accuracy and efficiency.
• Integrating advanced AI techniques can potentially improve diagnostic capabilities.

Purpose of the Study:

• To develop a hybrid artificial intelligence (AI) approach for improved Down Syndrome (DS) risk prediction in first trimester prenatal screening.
• To enhance the accuracy, efficiency, and reliability of DS risk assessment using novel data transformation and deep learning techniques.
• To explore the utility of transformer-based deep learning models for feature extraction from patient data represented as images.

Main Methods:

• Transformed 1D patient data (nuchal translucency, hCG, PAPP-A) into 2D Aztec barcode images.
• Employed transformer architectures (DeiT3, MaxViT, Swin) for feature extraction from barcode images.
• Utilized mRMR and RelieF for feature selection, followed by classification with Bagged Trees and Naive Bayes.

Main Results:

• Achieved up to 100% classification accuracy using Naive Bayes with selected features.
• Demonstrated significant reduction in hardware and processing demands through optimized feature selection.
• The hybrid AI approach proved effective in identifying high-risk cases with high precision.

Conclusions:

• The proposed hybrid AI method presents a promising, resource-efficient solution for Down Syndrome risk assessment in early prenatal screening.
• The image-based data transformation and deep learning feature extraction show potential for clinical application.
• Further validation in diverse clinical settings is recommended to confirm the generalizability and robustness of the approach.