Introduction
Prenatal diagnostics are fundamental to ensuring maternal-fetal health, yet significant barriers persist in delivering these services to underserved populations. Conventional diagnostic methods—including sonography, biochemical screenings, and genetic analysis—often require expensive equipment and specialized personnel typically concentrated in urban medical centers. This "last-mile" problem in healthcare leads to delayed diagnoses and preventable complications for expectant mothers in rural or understaffed regions.
To address these challenges, this research proposes a remote monitoring framework that integrates Internet of Things (IoT) technologies with Decision Tree Algorithms (DTA). By shifting from intermittent clinical "snapshots" to continuous, real-time data collection, the system aims to provide an automated safety net capable of identifying physiological abnormalities as they occur.
Methodology and System Architecture
The proposed diagnostic framework is built upon a multi-layered architecture designed to manage the flow of data from initial sensor acquisition to final clinical interpretation.
IoT Deployment and Data Acquisition
The system utilizes a comprehensive suite of sensors to monitor both maternal vitals and environmental factors. This includes consumer wearables, medical-grade patches, and home-based ambient sensors. The deployment strategy ensures that critical parameters such as heart rate, fetal movement, and oxygen saturation are tracked continuously.
Data Preprocessing
Raw IoT data is inherently noisy due to transmission errors and signal interference. To ensure data integrity, the framework employs several signal processing techniques:
- Z-score Normalization: Used to standardize signals and mitigate the impact of outliers.
- Median Filtering: Applied to smooth spikes caused by intermittent transmission errors.
- Independent Component Analysis (ICA): Utilized to resolve "intermixing" errors where different physiological signals overlap.
- Mean Imputation: Employed to handle missing data points within the stream.
The Decision Tree Model
The core logic of the system relies on the Classification and Regression Tree (CART) model. Unlike "black box" neural networks, decision trees offer high interpretability, which is vital in medical contexts. The algorithm partitions data based on "Information Gain," creating a hierarchical set of rules that clinicians can validate. For instance, a rule might trigger a check for oxygen saturation only if maternal heart rate and glucose levels exceed specific thresholds.
Datasets and Validation
The study was validated using a substantial dataset of over 100,000 individual records collected from pregnant women over a two-year period. For detailed analysis and fine-tuning, a subset of 5,000 records was randomly selected. The model's performance was further tested against a 1,000-record database of prenatal health records to compare its efficacy against classical statistical methods. The data was split into 80% for training and 20% for testing, with a portion of the training set reserved for hyperparameter optimization.
Results and Discussion
The DTA-driven IoT system demonstrated a significant improvement over traditional diagnostic paradigms. Key performance metrics include:
- Diagnostic Identification: The model achieved a 95% success rate in identifying potential health complications, compared to 85% for classical statistical analysis.
- Model Accuracy: In comparative testing, the DTA achieved an accuracy of 87.5%, a precision of 87.5%, and an F1-score of 82.4%.
- Error Reduction: The system successfully reduced false positives by 20% and false negatives by 15%, which is critical for minimizing patient anxiety and preventing overlooked emergencies.
While models like Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) showed higher theoretical Area-Under-the-Curve (AUC) scores in some tests, the DTA was preferred for its superior balance of accuracy and clinical interpretability.
Conclusion and Future Work
This research highlights the potential of combining IoT and interpretable machine learning to bridge the maternal health gap. By providing continuous monitoring and explainable diagnostic flags, the system offers a scalable solution for remote prenatal care.
Future work will focus on addressing infrastructure limitations, such as the reliance on stable Wi-Fi in remote areas, and enhancing data privacy protocols. Additionally, the authors plan to explore ensemble methods like Random Forests to further increase accuracy while maintaining the transparency of the decision-making process.
Summary updated