scripts/model_dev_pred_fatty_analysis.py

#!/usr/bin/env python3
"""
Calculate muscle fat percentages from CT images using model predictions
"""

import os
import re
from pathlib import Path
import numpy as np # type: ignore
import pandas as pd # type: ignore
import nibabel as nib # type: ignore

root_100_120 = Path("../100-120")
output_csv_100_120 = Path("../fatty_data/model_pred_dev.csv")

fat_hu_thresh = -20

muscle_labels = {
    1: "psoas",
    2: "quadratus_lumborum",
    3: "paraspinal",
    4: "latissimus_dorsi",
    5: "iliacus",
    6: "rectus_femoris",
    7: "vastus",
    8: "rhomboid",
    9: "trapezius",
}

def load_image_and_label_100_120(case_id: int, root_dir: Path):
    """Load CT image and model prediction label file for cases 100-120."""

    img_path = root_dir / "images_100-120" / f"{case_id}_0000.nii.gz"
    if not img_path.exists():
        return None, None

    lab_path = root_dir / "label_9_muscles_model_pred" / f"{case_id}.nii.gz"
    if not lab_path.exists():
        return None, None
    
    try:
        img = nib.load(str(img_path))  # type: ignore
        lab = nib.load(str(lab_path))  # type: ignore
        return img.get_fdata(), lab.get_fdata()
    except Exception as e:
        print(f"Error loading case {case_id}: {e}")
        return None, None

def calculate_fat_percentages(img_arr: np.ndarray, lab_arr: np.ndarray):
    """Calculate fat percentage (HU <= -20) for all 9 muscle labels."""
    fat_mask = img_arr <= fat_hu_thresh
    fat_percentages = {}
    
    for label_id, muscle_name in muscle_labels.items():
        muscle_mask = (lab_arr == label_id)
        total_voxels = int(np.count_nonzero(muscle_mask))  # type: ignore
        
        if total_voxels == 0:
            fat_pct = 0.0
        else:
            fat_voxels = int(np.count_nonzero(fat_mask & muscle_mask))  # type: ignore
            fat_pct = (fat_voxels / total_voxels) * 100.0
        
        fat_percentages[f"{muscle_name}_fat_pct"] = round(fat_pct, 2)
    
    return fat_percentages

def save_mean_std_stats(df, fat_cols, output_path):
    """Save mean ± SD statistics to a separate CSV file."""
    stats_data = []
    
    for col in fat_cols:
        values = df[col].values
        mean_val = np.mean(values)  # type: ignore
        std_val = np.std(values)  # type: ignore
        
        stats_data.append({
            'muscle': col.replace('_fat_pct', ''),
            'mean': round(mean_val, 2),
            'std': round(std_val, 2),
            'mean_std': f"{mean_val:.2f} ± {std_val:.2f}"
        })
    
    stats_df = pd.DataFrame(stats_data)
    stats_df.to_csv(output_path, index=False)
    print(f"Mean ± SD statistics saved to: {output_path}")

def process_100_120_dataset():
    """Process cases 100-120 using model predictions and save to model_pred_dev.csv"""
    print("="*60)
    print("PROCESSING CASES 100-120 WITH MODEL PREDICTIONS")
    print("="*60)
    
    rows = []

    print("Processing cases 100-120 with model predictions...")
    for case_id in range(100, 121):
        img_arr, lab_arr = load_image_and_label_100_120(case_id, root_100_120)
        if img_arr is not None and lab_arr is not None:
            fat_percentages = calculate_fat_percentages(img_arr, lab_arr)
            record = {"case_id": case_id, "dataset": "100-120_model_pred"}
            record.update(fat_percentages)
            rows.append(record)
            print(f"Case {case_id}: {[v for v in fat_percentages.values()][:3]}... %")
        else:
            print(f"Case {case_id}: Failed to load")
    
    if rows:
        df = pd.DataFrame(rows)

        fat_cols = [col for col in df.columns if col.endswith("_fat_pct")]

        fat_means = df[fat_cols].mean().round(2)
        fat_stds = df[fat_cols].std().round(2)

        summary_row = {"case_id": "Mean ± SD"}
        for col in fat_cols:
            summary_row[col] = f"{fat_means[col]:.2f} ± {fat_stds[col]:.2f}"

        summary_df = pd.DataFrame([summary_row])
        df_with_summary = pd.concat([df, summary_df], ignore_index=True)  # type: ignore

        output_csv_100_120.parent.mkdir(parents=True, exist_ok=True)

        df_with_summary.to_csv(output_csv_100_120, index=False)
        print(f"\nSaved {len(rows)} cases to {output_csv_100_120}")

        print(f"\nFat Percentage Summary (100-120 Model Predictions):")
        for col in fat_cols:
            values = df[col].values
            mean_val = np.mean(values)  # type: ignore
            std_val = np.std(values)  # type: ignore
            print(f"{col}: {mean_val:.2f} ± {std_val:.2f} %")

        save_mean_std_stats(df, fat_cols, output_csv_100_120.parent / "model_pred_dev_mean_std.csv")
    else:
        print("No data processed for 100-120!")

def main():
    """Main function to process the development dataset with model predictions."""
    print("FATTY PERCENTAGE ANALYSIS - MODEL PREDICTIONS")
    print("Computing fat percentage (HU <= -20) for 9 muscle labels using model predictions")
    print("="*60)

    process_100_120_dataset()
    
    print("\n" + "="*60)
    print("ANALYSIS COMPLETE")
    print("="*60)
    print(f"Results saved to:")
    print(f"  - {output_csv_100_120} (cases 100-120 with model predictions)")
    print(f"  - {output_csv_100_120.parent / 'model_pred_dev_mean_std.csv'} (summary statistics)")

if __name__ == "__main__":
    main()