🧬 Ovarian Cyst Predictor & Treatment Recommendation System

An AI-powered clinical decision support tool that predicts ovarian cyst behavior and recommends treatment pathways using advanced machine learning techniques. Built for early detection and support in PCOS-related and gynecological care. 🚀 Project Goals

Accurately predict the likelihood of cyst growth or malignancy using ML models trained on clinical and demographic data.

Recommend treatment plans (observation, medication, or surgery) aligned with evidence-based guidelines.

Export downloadable Excel reports with prediction confidence and clinical red flags.

Future features include real-time medication/surgical inventory tracking, cost estimation, and financing support.

🧠 Recent Milestone

✅ Achieved 90% prediction accuracy using Random Forest and XGBoost classifiers, improving performance and reliability over the initial Decision Tree model.

🛠️ Tech Stack

Python: Pandas, scikit-learn, NumPy, XGBoost

Modeling: Random Forest, XGBoost (with hyperparameter tuning)

Reporting: ExcelWriter for generating clinical reports

Deployment: Streamlit (web-based demo)

Version Control: Git & GitHub

📊 Sample Output (Excel Report)

The tool exports a rich Excel report combining clinical input data, derived features, and predictions from multiple machine learning models. Below is a sample snapshot of key fields:

Patient ID	Age	Cyst Size (cm)	CA-125	Symptoms	Predicted Behavior	Recommended Treatment	Confidence	Clinical Flag
OC-1000	52	3.2	19	Pelvic pain, Nausea, Bloating	Unstable	Observation	0.87	OK
OC-1001	62	7.9	111	Bloating	Unstable	Surgery	0.91	OK
OC-1002	59	2.2	123	Pelvic pain, Irregular periods	Unstable	Medication	0.84	OK
OC-1003	64	5.5	116	Nausea, Irregular periods	Unstable	Surgery	0.95	OK

🧠 Model Architecture & Prediction Logic

Feature Engineering includes clinical ratios like CA125_Size_Ratio, Symptom_Age_Ratio, and binary flags for post-menopausal risk.

Random Forest & XGBoost Models both generate predictions with probability scores.

Ensemble Logic:

    If both models agree → prediction adopted

    If models differ → select based on highest confidence score

Treatment Recommendation is rule-based and layered over the model output.

✅ Current Model Accuracy: 90%, validated on structured clinical cases. 🧪 Possible Next Steps

Incorporate patient history and hormonal profiles for deeper personalization

Add NLP pipeline for symptom description input

Partner with clinicians for field testing and feedback

📄 License

MIT License — Open source and free to use or adapt with attribution. 👩🏽‍💻 Developed by Wendy Wanjiru

MSc Software Engineering | AI for Healthcare Advocate

🔗 Connect on LinkedIn: https://www.linkedin.com/in/wendy-waweru18/ 🔗 Try it now! : https://wendyshiro-ovarian-cyst-predictor-streamlit-app-demo-sbp02b.streamlit.app/