scripts/pearson_analysis.py

#!/usr/bin/env python3
"""
Pearson correlation analysis for muscle fat vs Cobb angles
"""

import pandas as pd # type: ignore
import numpy as np # type: ignore
from scipy.stats import pearsonr  # type: ignore
from pathlib import Path

dev_fat_file = "../fatty_data/dev_fat.csv"
dev_cobb_file = "../cobb_angles/dev_cobb.csv"
dev_output_dir = "../pearson_correlation/dev_cobb_corr"

# Model prediction data
dev_model_pred_fat_file = "../fatty_data/model_pred_dev.csv"
dev_model_pred_output_dir = "../pearson_correlation/dev_model_cobb_corr"

test_fat_file = "../fatty_data/test_fat.csv"
test_cobb_file = "../cobb_angles/test_cobb.csv"
test_output_dir = "../pearson_correlation/test_cobb_corr"

muscle_names = [
    "psoas", "quadratus_lumborum", "paraspinal", "latissimus_dorsi", 
    "iliacus", "rectus_femoris", "vastus", "rhomboid", "trapezius"
]

def load_dev_data():
    """Load development dataset (100-120)."""
    print("Loading development dataset...")
    
    fat_df = pd.read_csv(dev_fat_file)
    print(f"Fatty data loaded: {len(fat_df)} cases")
    
    cobb_df = pd.read_csv(dev_cobb_file, sep='\t', header=None)  # type: ignore
    print(f"Cobb data loaded: {len(cobb_df)} cases")
    
    fat_df = fat_df[fat_df['case_id'] != 'Mean ± SD'].copy()
    fat_df = fat_df[pd.to_numeric(fat_df['case_id'], errors='coerce').notna()]  # type: ignore
    fat_df['case_id'] = fat_df['case_id'].astype(int)
    
    n_cases = min(len(cobb_df), len(fat_df))
    print(f"Using {n_cases} cases for development correlation analysis")
    
    cobb_values = cobb_df.iloc[:n_cases].values
    cobb_aligned = np.mean(cobb_values, axis=1)  # type: ignore
    fat_aligned = fat_df.iloc[:n_cases]
    
    print(f"Cobb angles range: {cobb_aligned.min():.1f} to {cobb_aligned.max():.1f}")
    
    return cobb_aligned, fat_aligned, n_cases

def load_dev_model_pred_data():
    """Load development dataset with model predictions (100-120)."""
    print("Loading development dataset with model predictions...")
    
    fat_df = pd.read_csv(dev_model_pred_fat_file)
    print(f"Model prediction fatty data loaded: {len(fat_df)} cases")
    
    cobb_df = pd.read_csv(dev_cobb_file, sep='\t', header=None)  # type: ignore
    print(f"Cobb data loaded: {len(cobb_df)} cases")
    
    fat_df = fat_df[fat_df['case_id'] != 'Mean ± SD'].copy()
    fat_df = fat_df[pd.to_numeric(fat_df['case_id'], errors='coerce').notna()]  # type: ignore
    fat_df['case_id'] = fat_df['case_id'].astype(int)
    
    n_cases = min(len(cobb_df), len(fat_df))
    print(f"Using {n_cases} cases for model prediction correlation analysis")
    
    cobb_values = cobb_df.iloc[:n_cases].values
    cobb_aligned = np.mean(cobb_values, axis=1)  # type: ignore
    fat_aligned = fat_df.iloc[:n_cases]
    
    print(f"Cobb angles range: {cobb_aligned.min():.1f} to {cobb_aligned.max():.1f}")
    
    return cobb_aligned, fat_aligned, n_cases

def load_test_data():
    """Load test dataset (251-500)."""
    print("Loading test dataset...")
    
    fat_df = pd.read_csv(test_fat_file)
    print(f"Fatty data loaded: {len(fat_df)} cases")
    
    cobb_df = pd.read_csv(test_cobb_file, header=None, names=['cobb_angle'])  # type: ignore
    print(f"Cobb data loaded: {len(cobb_df)} cases")

    fat_df = fat_df[fat_df['case_id'] != 'Mean ± SD'].copy()
    fat_df = fat_df[pd.to_numeric(fat_df['case_id'], errors='coerce').notna()]  # type: ignore
    fat_df['case_id'] = fat_df['case_id'].astype(int)

    n_cases = min(len(cobb_df), len(fat_df))
    print(f"Using {n_cases} cases for test correlation analysis")

    cobb_aligned = cobb_df.iloc[:n_cases]['cobb_angle'].values
    fat_aligned = fat_df.iloc[:n_cases]
    
    print(f"Cobb angles range: {cobb_aligned.min():.1f} to {cobb_aligned.max():.1f}")
    
    return cobb_aligned, fat_aligned, n_cases

def calculate_correlations(cobb_angles, fat_data, dataset_name):
    """Calculate Pearson correlations between Cobb angles and fatty percentages."""
    
    print(f"\nCalculating correlations for {dataset_name} dataset...")
    results = []
    
    for muscle in muscle_names:
        fat_col = f"{muscle}_fat_pct"
        
        if fat_col in fat_data.columns:
            fat_percentages = pd.to_numeric(fat_data[fat_col], errors='coerce').values  # type: ignore
            valid_indices = ~np.isnan(fat_percentages)  # type: ignore
            if not np.any(valid_indices):
                print(f"Warning: No valid data for {muscle}")
                continue

            cobb_filtered = cobb_angles[valid_indices]
            fat_filtered = fat_percentages[valid_indices]

            correlation, p_value = pearsonr(cobb_filtered, fat_filtered)  # type: ignore

            results.append({
                'Muscle': muscle,
                'Correlation': round(correlation, 4),
                'P_Value': round(p_value, 4),
                'N_Cases': len(cobb_filtered)
            })
            
            print(f"{muscle}: r = {correlation:.4f}, p = {p_value:.4f}")
        else:
            print(f"Warning: Column {fat_col} not found in fatty data")
    
    return results

def save_results(results, output_dir, dataset_name):
    """Save correlation results to CSV."""

    output_dir.mkdir(parents=True, exist_ok=True)

    results_df = pd.DataFrame(results)
    output_file = output_dir / "fatty_atrophy_thoracic_correlations.csv"
    results_df.to_csv(output_file, index=False)
    print(f"\nResults saved to: {output_file}")

    print(f"\n=== {dataset_name.upper()} CORRELATION ANALYSIS SUMMARY ===")
    print(f"Total muscles analyzed: {len(results)}")
    print(f"Cases used: {results[0]['N_Cases'] if results else 'N/A'}")

    if results:
        strongest_positive = max(results, key=lambda x: x['Correlation'])
        strongest_negative = min(results, key=lambda x: x['Correlation'])
        
        print(f"\nStrongest positive correlation: {strongest_positive['Muscle']} (r = {strongest_positive['Correlation']})")
        print(f"Strongest negative correlation: {strongest_negative['Muscle']} (r = {strongest_negative['Correlation']})")

        significant = [r for r in results if r['P_Value'] < 0.05]
        print(f"Significant correlations (p < 0.05): {len(significant)}/{len(results)}")

def main():
    """Main function to run correlation analysis for both datasets."""
    print("=== PEARSON CORRELATION ANALYSIS ===")
    print("Cobb angles vs Fatty percentages")
    print("="*50)
    
    try:
        print("\n" + "="*50)
        print("DEVELOPMENT DATASET ANALYSIS (100-120) - MANUAL LABELS")
        print("="*50)
        cobb_dev, fat_dev, n_dev = load_dev_data()
        results_dev = calculate_correlations(cobb_dev, fat_dev, "Development")
        save_results(results_dev, Path(dev_output_dir), "Development")

        print("\n" + "="*50)
        print("DEVELOPMENT DATASET ANALYSIS (100-120) - MODEL PREDICTIONS")
        print("="*50)
        cobb_dev_model, fat_dev_model, n_dev_model = load_dev_model_pred_data()
        results_dev_model = calculate_correlations(cobb_dev_model, fat_dev_model, "Development Model Predictions")
        save_results(results_dev_model, Path(dev_model_pred_output_dir), "Development Model Predictions")

        print("\n" + "="*50)
        print("TEST DATASET ANALYSIS (251-500)")
        print("="*50)
        cobb_test, fat_test, n_test = load_test_data()
        results_test = calculate_correlations(cobb_test, fat_test, "Test")
        save_results(results_test, Path(test_output_dir), "Test")
        
        print("\n" + "="*50)
        print("ANALYSIS COMPLETE")
        print("="*50)
        print(f"Development dataset (manual): {n_dev} cases analyzed")
        print(f"Development dataset (model): {n_dev_model} cases analyzed")
        print(f"Test dataset: {n_test} cases analyzed")
        print(f"Results saved to:")
        print(f"  - {dev_output_dir}/fatty_atrophy_thoracic_correlations.csv")
        print(f"  - {dev_model_pred_output_dir}/fatty_atrophy_thoracic_correlations.csv")
        print(f"  - {test_output_dir}/fatty_atrophy_thoracic_correlations.csv")
        
    except Exception as e:
        print(f"Error during analysis: {e}")
        return False
    
    return True

if __name__ == "__main__":
    main()