Inspiration

FibroPred was inspired by the urgent need to improve prognostic predictions for pulmonary fibrosis, a life-threatening disease with uncertain outcomes. Seeing the challenges clinicians face due to the lack of precise prediction tools motivated us to create a solution that could guide treatment decisions from the start.

What it does

FibroPred is an AI-powered predictive tool that forecasts the progression of pulmonary fibrosis from the point of diagnosis. It leverages machine learning algorithms to provide personalized risk assessments and treatment recommendations, ensuring better-informed clinical decisions.

How we built it

  • Data Collection & Preprocessing: We gathered anonymized patient data, ensuring adherence to medical privacy standards. Relevant clinical features like age, lung function, and genetic markers were selected and processed.
  • Model Development: We used XGBoost, optimizing it for three stages: diagnosis, one-year follow-up, and two-year follow-up. We integrated SHAP-based explainability to make model predictions transparent and interpretable.
  • Web Interface: We developed an intuitive web platform featuring data input, real-time predictions, and model retraining capabilities.
  • Testing & Validation: We tested the model using historical patient data, refining it based on expert feedback.

Challenges we ran into

-Understanding Relevant Variables: We needed to fully understand and select relevant variables by studying the medical context, ensuring our model captured essential clinical factors.

  • Model Accuracy vs. Interpretability: Balancing the model's predictive power with the need for transparent, explainable results.
  • User Experience: Creating an interface simple enough for healthcare professionals to use efficiently.

Accomplishments that we're proud of

  • Understanding Medical Context: We deepened our knowledge of the background and context of patients suffering from pulmonary fibrosis, gaining insights into the diagnostic process doctors follow.
  • Model Explainability: Successfully integrating SHAP for clear, interpretable predictions.

What we learned

  • Interdisciplinary Collaboration: Working closely with medical experts was crucial for clinical relevance.
  • Data Processing: We improved our understanding of handling complex medical datasets securely.
  • Model Development: We learned how to balance accuracy with transparency to build trust in our AI system.

What's next for FibroPred: Precisió Predictiva en l'Atenció Pulmonar.

  • Including More Patient Data: We plan to incorporate data from many more patients to improve accuracy, especially at the first stage of diagnosis.
  • Adopting Advanced Architectures: We aim to transition to more complex deep learning models capable of identifying intricate relationships within the data.
  • Seamless Integration: Supporting real-time data updates and enabling integration into hospital systems worldwide to set a new standard for personalized pulmonary care through predictive precision and explainability.

Built With

Share this project:

Updates