PrediCT GSoC 2026 — Project 2: Radiomics Feature Extraction & Calcium Phenotyping

Applicant: Laura Pinero Roig — laurapineroroig@gmail.com Project: PREDICT2 — Radiomics Feature Extraction and Calcium Phenotype Discovery Organization: ML4Sci

---

Setup

python3 -m venv predict2_env
source predict2_env/bin/activate
pip install -r requirements.txt

Data

Follow the instructions in the PrediCT repo to download and preprocess the Stanford COCA dataset. Place processed NIfTI files in data/processed/ with structure:

data/processed/
    patient_001/
        image.nii.gz
        mask.nii.gz
    patient_002/
        ...

Running the pipeline

# 1. Dataset statistics (common task)
python src/preprocess.py data/processed

# 2. Agatston scores
python src/agatston.py data/processed

# 3. Radiomics feature extraction (20-30 patients)
python src/feature_extraction.py data/processed

# 4. Statistical analysis + visualizations
cd notebooks && python analysis.py

Outputs

All outputs are saved to outputs/:

• dataset_statistics.csv — per-patient CT statistics
• agatston_scores.csv — scores and categories per patient
• radiomics_features.csv — PyRadiomics feature matrix
• spearman_correlations.csv — feature-Agatston correlations with p-values
• kruskal_wallis_results.csv — Kruskal-Wallis test across categories
• correlation_matrix.png — heatmap of top features
• feature_vs_agatston.png — scatter plots
• tsne_plot.png — t-SNE colored by Agatston category
• clustering.png — K-Means clusters vs Agatston categories