correlation viz

scripts/correlation_plot.py

@@ -0,0 +1,331 @@
+#!/usr/bin/env python3
+"""
+Create correlation plots for muscle fat vs Cobb angles
+"""
+
+import pandas as pd # type: ignore
+import numpy as np # type: ignore
+import matplotlib.pyplot as plt # type: ignore
+import seaborn as sns # type: ignore
+from pathlib import Path
+from scipy.stats import pearsonr  # type: ignore
+import argparse
+
+plt.style.use('seaborn-v0_8')
+sns.set_palette("husl")
+
+dev_correlation_csv = Path("../pearson_correlation/dev_cobb_corr/fatty_atrophy_thoracic_correlations.csv")
+test_correlation_csv = Path("../pearson_correlation/test_cobb_corr/fatty_atrophy_thoracic_correlations.csv")
+
+dev_fatty_csv = Path("../fatty_data/dev_fat.csv")
+test_fatty_csv = Path("../fatty_data/test_fat.csv")
+
+dev_cobb_csv = Path("../cobb_angles/dev_cobb.csv")
+test_cobb_csv = Path("../cobb_angles/test_cobb.csv")
+
+def load_correlation_data(dataset="dev"):
+    """Load the correlation data from the CSV file."""
+    if dataset == "dev":
+        csv_path = dev_correlation_csv
+    else:
+        csv_path = test_correlation_csv
+    
+    if not csv_path.exists():
+        print(f"Error: Correlation file not found at {csv_path}")
+        return None
+    
+    df = pd.read_csv(csv_path)
+    print(f"Loaded correlation data: {len(df)} muscles")
+    return df
+
+def create_dev_correlation_scatter(df):
+    """Create a scatter plot for development dataset (100-120)."""
+
+    fatty_df = pd.read_csv(dev_fatty_csv)
+    manual_cobb_df = pd.read_csv(dev_cobb_csv, sep='\t', header=None)  # type: ignore
+
+    thor_avg = np.round(manual_cobb_df.mean(axis=1)).astype(int)
+
+    fatty_manual = fatty_df[fatty_df['case_id'].str.isdigit()].copy()
+    fatty_manual['case_id'] = pd.to_numeric(fatty_manual['case_id'])  # type: ignore
+
+    fig, ax = plt.subplots(figsize=(14, 8))  # type: ignore
+    fig.patch.set_facecolor('#f8f9fa')
+    ax.set_facecolor('#ffffff')
+
+    muscle_cols = [col for col in fatty_manual.columns if col.endswith('_fat_pct')]
+
+    trapezius_col = None
+    for col in muscle_cols:
+        if 'trapezius' in col.lower():
+            trapezius_col = col
+            break
+    
+    if trapezius_col is None:
+        print("Trapezius muscle not found in the data")
+        return None, None
+
+    colors = ['#1f77b4']  # Blue color for Trapezius
+
+    col = trapezius_col
+
+    muscle_data = pd.to_numeric(fatty_manual[col], errors='coerce').values  # type: ignore
+    cobb_data = thor_avg[:len(muscle_data)]
+
+    valid_mask = ~(np.isnan(muscle_data) | np.isnan(cobb_data))  # type: ignore
+    muscle_clean = muscle_data[valid_mask]
+    cobb_clean = cobb_data[valid_mask]
+    
+    if len(muscle_clean) > 1:  # Need at least 2 points
+
+        muscle_name = col.replace('_fat_pct', '').replace('_', ' ').title()
+
+        ax.scatter(muscle_clean, cobb_clean, color=colors[0],   # type: ignore
+                  label=muscle_name, s=60, alpha=0.7, edgecolors='black', linewidth=0.5)
+
+        z = np.polyfit(muscle_clean, cobb_clean, 1)
+        p = np.poly1d(z)
+        ax.plot(muscle_clean, p(muscle_clean), color=colors[0],   # type: ignore
+               linestyle='-', alpha=0.8, linewidth=2)
+
+        muscle_name = col.replace('_fat_pct', '')
+        correlation_row = df[df['Muscle'] == muscle_name]
+        if not correlation_row.empty:
+            r = correlation_row['Correlation'].iloc[0]
+            p_val = correlation_row['P_Value'].iloc[0]
+        else:
+
+            r, p_val = pearsonr(muscle_clean, cobb_clean)  # type: ignore
+
+    ax.set_xlabel('Fat Percentage (%)', fontsize=12, fontweight='bold')
+    ax.set_ylabel('Thoracic Cobb Angle (deg)', fontsize=12, fontweight='bold')
+    ax.set_title('Trapezius Muscle Fat Percentage vs Thoracic Cobb Angle\n(n = 21 cases)', 
+                 fontsize=14, fontweight='bold', pad=20)
+
+    ax.grid(True, alpha=0.3, color='gray', linestyle='-', linewidth=0.5)
+
+    for spine in ax.spines.values():
+        spine.set_edgecolor('#333333')
+        spine.set_linewidth(1.5)
+
+    plt.tight_layout()
+
+    output_path = "../pearson_correlation/dev_cobb_corr/muscle_correlation_scatter.png"
+    plt.savefig(output_path, dpi=300, bbox_inches='tight', 
+                facecolor='white', edgecolor='none')
+    print(f"Saved correlation scatter plot to: {output_path}")
+    
+    return fig, ax
+
+def create_test_correlation_scatter(df):
+    """Create a scatter plot for test dataset (251-500)."""
+
+    fatty_df = pd.read_csv(test_fatty_csv)
+    cobb_df = pd.read_csv(test_cobb_csv, header=None, names=['cobb_angle'])  # type: ignore
+
+    fatty_df = fatty_df[fatty_df['case_id'] != 'Mean ± SD'].copy()
+    fatty_df = fatty_df[pd.to_numeric(fatty_df['case_id'], errors='coerce').notna()]  # type: ignore
+    fatty_df['case_id'] = fatty_df['case_id'].astype(int)
+
+    n_cases = min(len(cobb_df), len(fatty_df))
+    print(f"Using {n_cases} cases for test correlation analysis")
+
+    cobb_values = cobb_df.iloc[:n_cases, 0].values  # Get the first column as numpy array
+    fat_values = fatty_df.iloc[:n_cases]
+
+    trapezius_col = None
+    for col in fat_values.columns:
+        if 'trapezius' in col.lower() and col.endswith('_fat_pct'):
+            trapezius_col = col
+            break
+    
+    if trapezius_col is None:
+        print("Trapezius muscle not found in the test data")
+        return None, None
+
+    trapezius_data = pd.to_numeric(fat_values[trapezius_col], errors='coerce').values  # type: ignore
+    cobb_data = cobb_values
+
+    valid_mask = ~(np.isnan(trapezius_data) | np.isnan(cobb_data))  # type: ignore
+    trapezius_clean = trapezius_data[valid_mask]
+    cobb_clean = cobb_data[valid_mask]
+    
+    print(f"Valid data points: {len(trapezius_clean)}")
+    print(f"Trapezius range: {trapezius_clean.min():.2f} to {trapezius_clean.max():.2f}")
+    print(f"Cobb range: {cobb_clean.min():.1f} to {cobb_clean.max():.1f}")
+
+    fig, ax = plt.subplots(figsize=(14, 8))  # type: ignore
+    fig.patch.set_facecolor('#f8f9fa')
+    ax.set_facecolor('#ffffff')
+
+    ax.scatter(trapezius_clean, cobb_clean, color='#1f77b4', s=60, alpha=0.7,   # type: ignore
+              edgecolors='black', linewidth=0.5)
+
+    z = np.polyfit(trapezius_clean, cobb_clean, 1)
+    p = np.poly1d(z)
+    ax.plot(trapezius_clean, p(trapezius_clean), color='#1f77b4',   # type: ignore
+           linestyle='-', alpha=0.8, linewidth=2)
+
+    muscle_name = trapezius_col.replace('_fat_pct', '')
+    correlation_row = df[df['Muscle'] == muscle_name]
+    if not correlation_row.empty:
+        r = correlation_row['Correlation'].iloc[0]
+        p_val = correlation_row['P_Value'].iloc[0]
+    else:
+
+        r, p_val = pearsonr(trapezius_clean, cobb_clean)  # type: ignore
+
+    ax.set_xlabel('Trapezius Fat Percentage (%)', fontsize=12, fontweight='bold')
+    ax.set_ylabel('Thoracic Cobb Angle (deg)', fontsize=12, fontweight='bold')
+    ax.set_title(f'Trapezius Muscle Fat Percentage vs Thoracic Cobb Angle\n(n = {len(trapezius_clean)} cases)', 
+                 fontsize=14, fontweight='bold', pad=20)
+
+    ax.grid(True, alpha=0.3, color='gray', linestyle='-', linewidth=0.5)
+
+    for spine in ax.spines.values():
+        spine.set_edgecolor('#333333')
+        spine.set_linewidth(1.5)
+
+    plt.tight_layout()
+
+    output_path = "../pearson_correlation/test_cobb_corr/trapezius_fat_vs_thoracic_cobb_250_cases.png"
+    plt.savefig(output_path, dpi=300, bbox_inches='tight', 
+                facecolor='white', edgecolor='none')
+    print(f"Saved test correlation plot to: {output_path}")
+    
+    return fig, ax
+
+def create_aggregate_plot(df, dataset="dev"):
+    """Create a 3x3 aggregate plot showing all 9 muscles."""
+
+    if dataset == "dev":
+        fatty_df = pd.read_csv(dev_fatty_csv)
+        cobb_df = pd.read_csv(dev_cobb_csv, sep='\t', header=None)  # type: ignore
+
+        cobb_data = np.round(cobb_df.mean(axis=1)).astype(int)
+        n_cases = 21
+    else:
+        fatty_df = pd.read_csv(test_fatty_csv)
+        cobb_df = pd.read_csv(test_cobb_csv, header=None, names=['cobb_angle'])  # type: ignore
+        cobb_data = cobb_df.iloc[:, 0].values
+        n_cases = 250
+
+    fatty_df = fatty_df[fatty_df['case_id'] != 'Mean ± SD'].copy()
+    fatty_df = fatty_df[pd.to_numeric(fatty_df['case_id'], errors='coerce').notna()]  # type: ignore
+    fatty_df['case_id'] = fatty_df['case_id'].astype(int)
+
+    muscle_cols = [col for col in fatty_df.columns if col.endswith('_fat_pct')]
+
+    fig, axes = plt.subplots(3, 3, figsize=(18, 15))  # type: ignore
+    fig.patch.set_facecolor('#f8f9fa')
+
+    axes_flat = axes.flatten()
+
+    colors = ['#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd', 
+              '#8c564b', '#e377c2', '#7f7f7f', '#bcbd22']
+    
+    for i, col in enumerate(muscle_cols):
+        if i >= 9:  # Only plot first 9 muscles
+            break
+            
+        ax = axes_flat[i]
+        ax.set_facecolor('#ffffff')
+
+        muscle_data = pd.to_numeric(fatty_df[col], errors='coerce').values  # type: ignore
+        cobb_clean = cobb_data[:len(muscle_data)]
+
+        valid_mask = ~(np.isnan(muscle_data) | np.isnan(cobb_clean))  # type: ignore
+        muscle_clean = muscle_data[valid_mask]
+        cobb_clean = cobb_clean[valid_mask]
+        
+        if len(muscle_clean) > 1:
+
+            muscle_name = col.replace('_fat_pct', '').replace('_', ' ').title()
+
+            ax.scatter(muscle_clean, cobb_clean, color=colors[i],   # type: ignore
+                     s=40, alpha=0.7, edgecolors='black', linewidth=0.3)
+
+            if len(muscle_clean) > 1:
+                z = np.polyfit(muscle_clean, cobb_clean, 1)
+                p = np.poly1d(z)
+                ax.plot(muscle_clean, p(muscle_clean), color=colors[i],   # type: ignore
+                       linestyle='-', alpha=0.8, linewidth=1.5)
+
+            muscle_name_csv = col.replace('_fat_pct', '')
+            correlation_row = df[df['Muscle'] == muscle_name_csv]
+            if not correlation_row.empty:
+                r = correlation_row['Correlation'].iloc[0]
+                p_val = correlation_row['P_Value'].iloc[0]
+            else:
+                r, p_val = pearsonr(muscle_clean, cobb_clean)  # type: ignore
+
+            ax.set_title(f'{muscle_name}\nr = {r:.3f}', fontsize=10, fontweight='bold')
+
+        ax.set_xlabel('Fat %', fontsize=8)
+        ax.set_ylabel('Cobb Angle (deg)', fontsize=8)
+        ax.grid(True, alpha=0.3, linewidth=0.5)
+
+        for spine in ax.spines.values():
+            spine.set_edgecolor('#333333')
+            spine.set_linewidth(0.8)
+
+    for i in range(len(muscle_cols), 9):
+        axes_flat[i].set_visible(False)
+
+    plt.tight_layout()
+
+    output_path = f"../pearson_correlation/{dataset}_cobb_corr/aggregate_muscle_correlations.png"
+    plt.savefig(output_path, dpi=300, bbox_inches='tight', 
+                facecolor='white', edgecolor='none')
+    print(f"Saved aggregate plot to: {output_path}")
+    
+    return fig, axes
+
+def main():
+    """Main function to create correlation plots."""
+    parser = argparse.ArgumentParser(description='Create muscle correlation plots')
+    parser.add_argument('--dataset', choices=['dev', 'test', 'both'], default='both',
+                       help='Which dataset to plot (dev, test, or both)')
+    parser.add_argument('--plot-type', choices=['trapezius', 'aggregate', 'both'], default='both',
+                       help='Which plot type to create (trapezius, aggregate, or both)')
+    args = parser.parse_args()
+    
+    print("=== MUSCLE CORRELATION VISUALIZATION ===")
+    
+    if args.dataset in ['dev', 'both']:
+        print("\n=== DEVELOPMENT DATASET (100-120) ===")
+        df_dev = load_correlation_data("dev")
+        if df_dev is not None:
+            if args.plot_type in ['trapezius', 'both']:
+                fig1, ax1 = create_dev_correlation_scatter(df_dev)
+                if fig1 is not None:
+                    print("Development trapezius plot created successfully")
+                plt.show()
+            
+            if args.plot_type in ['aggregate', 'both']:
+                fig2, ax2 = create_aggregate_plot(df_dev, "dev")
+                if fig2 is not None:
+                    print("Development aggregate plot created successfully")
+                plt.show()
+    
+    if args.dataset in ['test', 'both']:
+        print("\n=== TEST DATASET (251-500) ===")
+        df_test = load_correlation_data("test")
+        if df_test is not None:
+            if args.plot_type in ['trapezius', 'both']:
+                fig3, ax3 = create_test_correlation_scatter(df_test)
+                if fig3 is not None:
+                    print("Test trapezius plot created successfully")
+                plt.show()
+            
+            if args.plot_type in ['aggregate', 'both']:
+                fig4, ax4 = create_aggregate_plot(df_test, "test")
+                if fig4 is not None:
+                    print("Test aggregate plot created successfully")
+                plt.show()
+    
+    print("\n=== VISUALIZATION COMPLETE ===")
+    print("Generated Pearson correlation scatter plots for muscle fat percentages vs thoracic Cobb angles")
+
+if __name__ == "__main__":
+    main()