Article Details

Abstract

Objective Macrosomia is a significant risk factor for adverse maternal and neonatal outcomes, yet traditional methods struggle to capture the dynamic trajectory of fetal growth, and existing Machine Learning models have difficulty in handling tabular time-series data. Methods This study proposes a macrosomia prediction model based on the iTransformer architecture integrated with dynamic temporal features. A total of 2 544 singleton pregnancies from 2020 to 2024 are retrospectively included. Growth rates of fetal biometric measurements are constructed as dynamic temporal features, dynamic dependencies among features are modeled through feature serialization and self-attention mechanisms, and a specificity-targeted optimization strategy is introduced to balance recall and misdiagnosis rates. Results Five-fold cross-validation shows that the model achieves a recall of 0.83±0.06, a specificity of 0.74±0.01, and a ROC-AUC of 0.85±0.03, and its recall outperforms baseline models such as LSTM, GRU, TCN, TabTransformer, Transformer and Random Forest. SHAP analysis identifies key predictive features including fetal abdominal circumference and femur length. Conclusion The proposed model effectively captures the dynamic patterns of prenatal examination time-series data under a controllable misdiagnosis rate, significantly improves the identification of macrosomia, and provides an explainable new method for dynamic prenatal risk assessment.

Keywords

macrosomia; iTransformer; temporal feature; self-attention mechanism

References

1. [1] Liu Q Y，Zhu S M，Zhao M，et al. Machine Learning Approaches for Predicting Fetal Macrosomia at Different Stages of Pregnancy: a Retrospective Study in China[J/OL].BMC Pregnancy and Childbirth，2025，25（1）：140.https://pmc.ncbi.nlm.nih.gov/articles/PMC11817776/.
2. [2] Li Y，Nair P，Lu X H，et al. Inferring multimodal latent topics from electronic health records[J/OL].Nature Communications，2020：2536.https://doi.org/10.1038/s41467-020-16378-3.
3. [3] Dodd J M，Louise J，Deussen A R，et al. Rethinking Causal Assumptions about Maternal BMI, Gestational Weight Gain, and Adverse Pregnancy Outcomes[J/OL].BMC Medicine，2024，22（1）：197.https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-024-03410-2.
4. [4] Chen T，Guestrin C. XGBoost: A Scalable Tree Boosting System[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining，2016：785-794.
5. [5] Sovio U，Murphy H R，Smith G C S. Accelerated Fetal Growth Prior to Diagnosis of Gestational Diabetes Mellitus: A Prospective Cohort Study of Nulliparous Women[J].Diabetes Care，2016，39（6）：982-987.
6. [6] Fernández A，García S，Herrera F，et al. SMOTE for Learning from Imbalanced Data: Progress and Challenges[J]. Journal of Artificial Intelligence Research，2018，61：863-905.
7. [7] Vaswani A，Shazeer N，Parmar N，et al. Attention is All You Need[C]//Advances in Neural Information Processing Systems 30 (NeurIPS 2017)，2017：1-11.
8. [8] Zeng A L，Chen M X，Zhang L，et al. Are Transformers Effective for Time Series Forecasting?[C]// Proceedings of the AAAI Conference on Artificial Intelligence，2023，37（9）：11121-11128.
9. [9] Liu Y，Hu T G，Zhang H R，et al. iTransformer: Inverted Transformers Are Effective for Time Series Forecasting[PP/OL]. arXiv:2310.06625 [cs.LG].https://doi.org/10.48550/arXiv.2310.06625.
10. [10] Zerveas G，Jayaraman S，Patel D，et al. A Transformer-based Framework for Multivariate Time Series Representation Learning[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining，2021：2114-2124.
11. [11] Lundberg S M，Lee S-I. A Unified Approach to Interpreting Model Predictions[C]//Advances in Neural Information Processing Systems 30 (NeurIPS 2017)，2017：1-10.
12. [12] Douzas G，Bacao F，Last F. Improving Imbalanced Learning Through a Heuristic Oversampling Method Based on k-means and SMOTE[J].Information Sciences，2018，465：1-20.
13. [13] Wang Z G，Yan W Z，Oates T. Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline[C]//2017 International Joint Conference on Neural Networks (IJCNN)，2017：1578-1585.

Funding

National College Student Innovation Training Program of Hangzhou Medical University in 2025 (2025130230013)

About the authors

Luo Jinlan (October 2003-), female, Han ethnicity, from Shaoxing, Zhejiang, undergraduate student, research interests include nursing and treatment of common obstetric and gynecological diseases; Xu Yihang (April 2005-), male, Han ethnicity, from Jiaxing, Zhejiang, undergraduate student, research interests include intelligent medical engineering; Corresponding author: Chen Wenjun (January 1989-), female, Han ethnicity, from Liaocheng, Shandong, lecturer, doctoral degree, research interests include integrated Chinese and Western medicine in the diagnosis and treatment of common obstetric and gynecological diseases.