CERVINet-DELTA

# CERVINet-DELTA: Multimodal Predictive Modeling for Dinoprostone-Induced Labor Outcomes > A transformer-driven, attention-guided framework for individualized labor induction outcome prediction and decision support, integrating clinical features, time series signals, and ultrasound imagery. ## 🧠 Overview CERVINet-DELTA is a multimodal machine learning system designed to predict and optimize clinical outcomes of labor induction using dinoprostone. Combining structured clinical data, dynamic uterine activity, and imaging features, it supports both high-accuracy forecasting and real-time decision-making during the labor process. This repository provides the full pipeline from data preprocessing, model architecture, training scripts, ablation studies, to figure generation and result visualization. --- ## 📊 Key Features - **Multimodal Fusion:** Bishop score components, uterine contractions, and ultrasound data- **Transformer-based Attention:** Cross-modal learning via dynamic attention weighting- **Time-series Prediction:** Accurate estimation of labor onset and delivery time- **Explainability:** Attention heatmaps and utility saliency for clinical interpretability- **Policy Optimization Layer (DELTA):** Counterfactual analysis and adaptive intervention --- ## 📁 Project Structure ```bash.├── data/ # Raw and preprocessed datasets├── models/ # CERVINet and DELTA architecture modules├── train/ # Training scripts and configuration├── inference/ # Evaluation, prediction and figure generation├── figures/ # Visual outputs for publications├── utils/ # Helper functions (metrics, loaders, samplers)├── data.py # Dataset class and loading functions├── requirements.txt # Python dependencies└── README.md # Project documentation