Scholarlink

<p>Down syndrome remains one of the most common chromosomal disorders worldwide, and accurate first-trimester risk stratification is essential for prenatal counseling and clinical decision-making. This study aimed to develop a deep learning model for comparison with the Fetal Medicine Foundation (FMF) algorithm for first-trimester Down syndrome risk classification in East Asian pregnant women. Using the FMF algorithm–derived risk classification as the reference standard, the original set of FMF input parameters was reduced to the five most informative indicators based on permutation importance analysis and clinical relevance. A retrospective cohort of 3,812 pregnancies from the first-trimester screening database at Taipei Chang Gung Memorial Hospital was used for model development and evaluation. The five retained indicators—maternal age, age-related background risk of Down syndrome, nuchal translucency thickness, PAPP-A, and free β-hCG—were identified as the most influential features, consistent with established clinical knowledge. Using these simplified indicators, the artificial neural network (ANN) achieved an area under the receiver operating characteristic curve (AUC) of 0.929 when evaluated against the original FMF algorithm–derived risk classification. The present findings should be interpreted as an FMF concordance study between the ANN model and standardized FMF-based screening. Overall, as a concordance study, the proposed ANN model using a reduced set of key features can approximate FMF-based risk classification. Further study is required to demonstrate the feasibility of the model in the real-world situation and to evaluate its clinical value using datasets with confirmed trisomy 21 outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning-based first-trimester down syndrome risk prediction in East Asian pregnant women using simplified clinical indicators

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • Yen-Tin Chen,
  • Yu-Shiang Lin

摘要

Down syndrome remains one of the most common chromosomal disorders worldwide, and accurate first-trimester risk stratification is essential for prenatal counseling and clinical decision-making. This study aimed to develop a deep learning model for comparison with the Fetal Medicine Foundation (FMF) algorithm for first-trimester Down syndrome risk classification in East Asian pregnant women. Using the FMF algorithm–derived risk classification as the reference standard, the original set of FMF input parameters was reduced to the five most informative indicators based on permutation importance analysis and clinical relevance. A retrospective cohort of 3,812 pregnancies from the first-trimester screening database at Taipei Chang Gung Memorial Hospital was used for model development and evaluation. The five retained indicators—maternal age, age-related background risk of Down syndrome, nuchal translucency thickness, PAPP-A, and free β-hCG—were identified as the most influential features, consistent with established clinical knowledge. Using these simplified indicators, the artificial neural network (ANN) achieved an area under the receiver operating characteristic curve (AUC) of 0.929 when evaluated against the original FMF algorithm–derived risk classification. The present findings should be interpreted as an FMF concordance study between the ANN model and standardized FMF-based screening. Overall, as a concordance study, the proposed ANN model using a reduced set of key features can approximate FMF-based risk classification. Further study is required to demonstrate the feasibility of the model in the real-world situation and to evaluate its clinical value using datasets with confirmed trisomy 21 outcomes.