CardioSense AI

Cardiovascular risk assessment platform built with Streamlit and scikit-learn.

Trains 6 ML models + a PyTorch neural network on clinical patient data, compares their performance, and provides per-patient explainability via SHAP values and rule-based clinical flags.

Project Structure

CardioSense-AI/
  app.py                      Main Streamlit entry point (config + styling)
  requirements.txt            Python dependencies
  pages/
    1_Overview.py             Landing page
    2_Train_Models.py         Dataset upload, training, evaluation
    3_Predict_Patient.py      Single patient risk prediction + SHAP
    4_Batch_Prediction.py     CSV batch predictions
    5_Dataset_Explorer.py     EDA visualisations
    6_Neural_Network.py       PyTorch CardioNet training
  src/
    data_loader.py            CSV loading and validation
    preprocessor.py           Feature engineering, scaling, SMOTE
    trainer.py                6 classical ML models
    deep_model.py             PyTorch neural network
    explainer.py              SHAP + clinical rule flags
    predictor.py              Inference wrapper
    plots.py                  Matplotlib / Seaborn visualisations
  data/
    heart.csv                 UCI Heart Disease dataset (303 records)
  models/                     Auto-created after training
  notebooks/
    pipeline_testing.ipynb    End-to-end pipeline notebook

Setup

git clone https://github.com/YOUR_USERNAME/CardioSense-AI.git
cd CardioSense-AI
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

On macOS with Apple Silicon, LightGBM requires OpenMP:

brew install libomp

Running

streamlit run app.py

Open http://localhost:8501. Navigate to Train Models, upload heart.csv, and train. Then use Predict Patient or Batch Prediction.

Dataset

Source: UCI Machine Learning Repository — Cleveland Heart Disease dataset (https://www.kaggle.com/datasets/johnsmith88/heart-disease-dataset)

303 patient records with 13 clinical features and a binary target (0 = no disease, 1 = disease).

Column	Description
age	Age in years
sex	1 = male, 0 = female
cp	Chest pain type (0-3)
trestbps	Resting blood pressure (mmHg)
chol	Serum cholesterol (mg/dl)
fbs	Fasting blood sugar > 120 mg/dl (1 = yes)
restecg	Resting ECG result (0, 1, 2)
thalach	Maximum heart rate achieved
exang	Exercise-induced angina (1 = yes)
oldpeak	ST depression during exercise
slope	Slope of peak exercise ST segment
ca	Number of major vessels coloured (0-3)
thal	Thalassemia type

Models

Model	Type
Logistic Regression	Linear
Decision Tree	Tree
Random Forest	Ensemble (bagging)
Gradient Boosting	Ensemble (boosting)
XGBoost	Gradient boosting
LightGBM	Gradient boosting
CardioNet	PyTorch feed-forward

All classical models support probability calibration (isotonic regression) and 5-fold stratified cross-validation. SMOTE is applied to the training set to handle class imbalance.

Explainability

• SHAP: TreeExplainer for tree-based models, KernelExplainer for others. Produces per-patient waterfall charts showing feature contributions.
• Clinical flags: Deterministic checks against known medical thresholds (e.g. cholesterol > 240, BP > 140).

Deployment

Push to GitHub, then deploy via:

• Streamlit Community Cloud: https://share.streamlit.io — set main file to app.py
• Hugging Face Spaces: Create a new Space with SDK = Streamlit

Disclaimer

This system is built for academic purposes. It has not been validated for clinical use and must not substitute advice from a qualified medical professional.