HopeCast: Breast Cancer Survival Prediction & Personalized Treatment Guidance

Overview

• HopeCast is an AI-powered web application that predicts breast cancer survival time using clinical, genetic, and mutation data.
• It also provides personalized treatment explanations by retrieving information from real medical literature.
• The project integrates classical ML (XGBoost), deep learning (DeepSurv), and a Retrieval-Augmented Generation (RAG) system for explainability.

Features

• Predicting survival using clinical, genetic and mutation data
• Comparing performance between Naive, XGBoost, and DeepSurv models
• Understanding predictions with RAG-powered explanations using real PubMed research
• User-friendly interface built with Streamlit

How It Works

1. Choose a dataset (Clinical, Genetic, or Mutation)
2. Enter patient details like age, tumor stage, and treatment history
3. The model predicts:
   • Median survival time
   • Survival probability at different months
4. A Retrieval-Augmented Generation (RAG) system is used to:
   • Search for relevant articles from PubMed
   • Extract meaningful medical content
   • Generate clear, personalized treatment explanations using a language model (via Grok API)

Model Results

Dataset	Naive Baseline	XGBoost	DeepSurv
Clinical	0.5000	0.6001	0.7135
Genetic	0.5000	0.6513	0.6730
Mutation	0.5000	0.5687	0.5611

Try the HopeCast Web App here:

https://hopecast-service-1004867270011.us-central1.run.app/

Setup Instructions to Run Locally

1. Clone the Repository

git clone https://github.com/johnjoel2001/HopeCast.git
cd HopeCast

2. Create and Activate a Virtual Environment

python -m venv venv
source venv/bin/activate

3. Install Requirements

pip install -r requirements.txt

4. Add Environment Variables ( Required if running on local machine )

Create a .env file in the project root with:

XAI_API_KEY=your_openai_or_grok_api_key
ENTREZ_EMAIL=your_email@example.com

Steps to Run all the Models Seperately

1. Download and Prepare Data

python scripts/make_dataset.py
python scripts/clean_data.py

2: Train Models

Naive Baseline

python scripts/naive_model.py

XGBoost

python scripts/non_deep_learning.py

DeepSurv

python scripts/deep_learning.py

Steps to Run the App

streamlit run app.py

Ethics & Responsibility

• HopeCast is a research tool, not a clinical decision system

• All data is anonymized and sourced from METABRIC (public)

• Explanations are generated from open-access PubMed literature

• Built to support, not replace medical professionals

Acknowledgements

• METABRIC Dataset (via Kaggle)

• DeepSurv (Katzman et al., 2018)

• xAI Grok API

• PubMed/NCBI Entrez

• FAISS, Lifelines, PyCox, Streamlit, and other open-source libraries