PHÁT TRIỂN VÀ ĐÁNH GIÁ NỘI BỘ PHẦN MỀM TRÍ TUỆ NHÂN TẠO HỖ TRỢ SÀNG LỌC TRƯỚC SINH HỘI CHỨNG EDWARDS VÀ PATAU

Abstract

Introduction: Edwards syndrome (trisomy 18) and Patau syndrome (trisomy 13) are severe chromosomal aneuploidies, while studies investigating the application of artificial intelligence (AI) to support screening for these conditions remain limited.

Objective: This study aimed to preliminarily evaluate the performance of an experimental AI-based software system for identifying the risk of Edwards and Patau syndromes, using fetal karyotyping as the reference standard.

Materials and methods: This retrospective study used 10,618 pregnancy records from three hospitals in Hanoi. Model development included 6,689 cases with karyotyping results: 272 Edwards, 62 Patau, and 6,355 unaffected cases. Data were split into 80% training and 20% validation sets using stratified sampling; SMOTE was applied only to training data. The software combined clinical, ultrasound, and biochemical data with expert knowledge and machine-learning algorithms. Random Forest achieved 100.0% sensitivity, 96.2% specificity, and 97.5% accuracy and was selected for integration. Results represent internal validation in a selected population undergoing karyotyping and may not generalize to routine screening.

Conclusion: The software demonstrated potential as a supportive tool for integrating clinical information and identifying pregnancies at increased risk of trisomy 18 and trisomy 13. External and prospective validation in independent screening populations is required before widespread clinical implementation.