Andres&Didac's team

The project

This project is a simple but effective tool to aid health specialists determine the pronostics pacients with endometrial cancer that fall into the exclusion diagnosis.

How it works

The core is a pre-trained model that will predict the risk of patients. On top of that model, a simple UI using Gradio is built into the Colab to be able to input data into the model and visualize results instantly.

Methodology

The methodology is centered on a Cox Proportional Hazards (CPH) Model :

1. Data Filtering: The dataset is first strictly filtered to include only NSMP cases (p53 Wild-Type and POLE Non-Mutated). python df_nsmp = df[ (df['p53_ihq'] == 1) & (df['mut_pole'] == 2) ].copy()
2. Model Training: A CPH Model with L2 regularization (penalizer=0.1) is trained using the lifelines library on key pathological and molecular features (FIGO Stage, Grade, LVSI, Myometrial Infiltration, and Beta-Catenin). This yields a Log-Hazard Ratio (Log-HR) for each factor. python cph_beta = CoxPHFitter(penalizer=0.1) cph_beta.fit(df_modelo, duration_col='diferencia_dias_reci_exit', event_col='recidiva')
3. Risk Score: The final NEST Score is the weighted sum of these Log-HR coefficients, determining the patient's individual risk category (Low, Intermediate, or High). $$\text{Risk Score} = \sum (\text{Log-HR}_i \times \text{Feature Value}_i)$$

Built With

• colab
• cox
• gradio
• python