A Hybrid CNN-LSTM Multi-Modal for Early Detection and Recurrence Prediction of Ovarian Cancer and Model Transparency using SHAP and LIME

Ovarian cancer is one of the deadliest cancers in women because it shows no clear symptoms in the early stages and often returns after treatment. Current detection methods are not strong enough and lack the ability to predict recurrence effectively. The research presents a robust hybrid machine learning-based approach, which integrates traditional Machine Learning(ML), Deep Learning (DL), and ensemble methods, in order to maximize early diagnosis, and predict tumor recurrence. More specifically, the studies closely examine three domains of data—gene expression, clinical record, and tissue images—to improve the predictive power of the approach. The hybrid approach employs advanced feature selection strategies that utilize PCA (Principal Component Analysis) and LASSO (Least Absolute Shrinkage and Selection Operator which the authors evaluate to determine and identify the best predictors of disease. We will apply Random Forest to patient data record analysis, CNNs (Convolutional Neural Networks) for image analysis of tissues, and LSTMs (Long Short-Term Memory) for gene sequences. An ensemble approach with models such as XGBoost (eXtreme Gradient Boosting) and stacking yields even more performance and robustness. Tests on real-world data show that this hybrid model performs significantly better than single methods, offering higher accuracy, precision, and fewer false results. To make the system’s decisions understandable for doctors, explainable AI tools like SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) are used. This study proves that combining different data types and smart techniques greatly improves the early detection and monitoring of ovarian cancer.