A recent research paper, "Prediction of high-risk pregnancy based on machine learning algorithms" by Xinyu Pi et al., highlights the growing role of machine learning (ML) in identifying high-risk pregnancies early—an area where timely interventions can truly make a difference. The study used real-world data from Bangladesh and found that multilayer perceptron (MLP) models achieved an impressive 91% accuracy in predicting high-risk pregnancies.

This paper offers a compelling example of AI’s potential—and also serves as a reminder of its limitations. In high-stakes areas like maternal healthcare, relying solely on machine-driven predictions isn’t enough. Human expertise must remain a core part of the process. Here’s why incorporating experts throughout the AI lifecycle is essential.
‍

What the Study Got Right: Leveraging AI in Healthcare with Expert Insight

The study showcased solid methodologies—data cleaning, class-balancing with SMOTE, and early stopping to avoid overfitting. The model used straightforward yet powerful features like age, blood pressure, glucose levels, temperature, and heart rate to predict pregnancy risk with notable accuracy.

Importantly, the study itself emphasizes a crucial point: AI should enhance, not replace, clinical expertise. By integrating domain knowledge into the development and application of machine learning models, predictions become not only technically sound but also contextually meaningful.
‍

Prediction Without Understanding Can Be Dangerous

While a 91% accuracy rate is remarkable, what about the remaining 9%? What if that includes a missed high-risk case or an overtreated low-risk patient?

In critical domains like obstetrics, interpretability and context matter. A model can identify statistical patterns, but only a trained clinician or expert can determine whether those patterns represent actionable insights. Domain experts are essential for validating training data, reviewing model outputs, and ensuring that decisions made by AI systems are safe and justifiable.
‍

The Value of Expert Context in Model Training

The study used data from Bangladesh, a region with unique healthcare practices and risk factors. An OB-GYN in Dhaka may interpret certain clinical indicators differently than one in Detroit, due to differences in healthcare infrastructure, environment, and prenatal care standards.

Without domain-specific input in the labeling and validation process, models risk being misaligned with the real-world settings they aim to serve. Experts help ensure that AI systems are sensitive to local context, which is critical for making accurate and ethical decisions.
‍

Beyond Accuracy: Accountability and Trust

Metrics like confusion matrices and ROC curves help validate technical performance, but they don’t capture the full picture—especially in healthcare. Trust is a key requirement. Clinicians must be able to understand how a model arrived at its conclusions, and patients deserve assurance that AI-supported recommendations are based on clinical best practices, not just code and computation.

Experts-in-the-loop play a vital role in making AI systems interpretable, ethical, and trustworthy. Human-AI collaboration is essential for ensuring that decisions align with professional standards and real-world patient needs.
‍

Final Thoughts

The research by Pi et al. makes a valuable contribution to maternal healthcare, showing the potential of ML in detecting high-risk pregnancies early. At Perle, we believe the next step is blending technical excellence with human expertise—especially in areas like data labeling, model evaluation, and real-time decision support.

We don’t just need models that predict better. We need systems that understand better. And that starts with keeping experts-in-the-loop.

References: Pi, X., Wang, J., Chu, L., Zhang, G., & Zhang, W. (2025). Prediction of high-risk pregnancy based on machine learning algorithms. Scientific Reports, 15, 15561. https://doi.org/10.1038/s41598-025-00450-3 

‍