CERVINet-DELTA: Multimodal Forecasting for Labor Induction

# CERVINet-DELTA: Multimodal Time-Series Framework for Labor Induction Prediction > A robust deep learning framework for individualized prediction of labor outcomes induced by dinoprostone, using multimodal fusion and temporal attention modeling. ## 🧠 Overview **CERVINet-DELTA** is an interpretable and personalized predictive system that models the dynamic process of labor induction. It fuses structured clinical features, ultrasound imagery, and time-series uterine activity using attention-enhanced neural networks to forecast key outcomes such as: - Time to active labor- Total time to delivery- Delivery mode (vaginal, assisted, cesarean)- Neonatal Apgar scores This framework supports **real-time decision making**, **risk stratification**, and **policy optimization** based on individualized clinical profiles. ## 📦 Features - ⚙️ **CERVINet**: Stochastic latent dynamics with encoder-decoder architecture and temporal GRU transitions.- 🧮 **DELTA**: Policy reasoning and utility-driven intervention module for dinoprostone titration.- 🧊 Multimodal data support: Time-series, tabular, and image modalities.- 📈 Evaluation across multiple labor datasets (Dinoprostone, Multimodal Labor, Attention-Weighted).- 🧪 Ablation studies on core modules. ## 📁 Folder Structure# CERVINet-DELTA：用于引产预测的多模态时间序列框架 > 本框架是一款鲁棒的深度学习模型，依托多模态融合与时序注意力建模技术，实现对地诺前列酮（dinoprostone）诱导引产结局的个性化精准预测。 ## 🧠 概述 **CERVINet-DELTA** 是一款具备可解释性的个性化预测系统，可对引产的动态过程进行建模。该系统通过注意力增强型神经网络，融合结构化临床特征、超声影像与子宫收缩时序信号，实现对以下关键结局的预测： - 活跃产程时长 - 总分娩时长 - 分娩方式（阴道分娩、辅助分娩、剖宫产） - 新生儿阿普加（Apgar）评分 本框架可基于患者个性化临床档案，支持**实时临床决策**、**风险分层**与**诊疗策略优化**。 ## 📦 核心特性 - ⚙️ **CERVINet**：采用编码器-解码器架构与时序门控循环单元（GRU）转换模块，实现随机潜在动力学建模。 - 🧮 **DELTA**：面向地诺前列酮剂量滴定的策略推理与效用导向干预模块。 - 🧊 支持多模态数据类型：涵盖时序数据、表格数据与影像模态。 - 📈 基于多组引产数据集开展模型评估（地诺前列酮数据集、多模态引产数据集、注意力加权数据集）。 - 🧪 针对核心模块开展消融实验研究。 ## 📁 文件夹结构