|
|
import pandas as pd
|
|
|
import numpy as np
|
|
|
import json
|
|
|
import argparse
|
|
|
import os
|
|
|
from datetime import datetime
|
|
|
import traceback
|
|
|
from io import StringIO
|
|
|
import sys
|
|
|
|
|
|
def evaluate_rsp_metrics(output_csv, ground_truth_csv):
|
|
|
log_output = StringIO()
|
|
|
sys_stdout = sys.stdout
|
|
|
sys.stdout = log_output
|
|
|
result = {
|
|
|
"Process": False,
|
|
|
"Result": False,
|
|
|
"TimePoint": datetime.now().isoformat(),
|
|
|
"comments": ""
|
|
|
}
|
|
|
|
|
|
try:
|
|
|
|
|
|
if not os.path.exists(output_csv):
|
|
|
print(f"Error: Output CSV file '{output_csv}' does not exist")
|
|
|
return result
|
|
|
if not os.path.exists(ground_truth_csv):
|
|
|
print(f"Error: Ground truth CSV file '{ground_truth_csv}' does not exist")
|
|
|
return result
|
|
|
if os.path.getsize(output_csv) == 0:
|
|
|
print(f"Error: Output CSV file '{output_csv}' is empty")
|
|
|
return result
|
|
|
if os.path.getsize(ground_truth_csv) == 0:
|
|
|
print(f"Error: Ground truth CSV file '{ground_truth_csv}' is empty")
|
|
|
return result
|
|
|
result["Process"] = True
|
|
|
|
|
|
|
|
|
output_df = pd.read_csv(output_csv)
|
|
|
gt_df = pd.read_csv(ground_truth_csv)
|
|
|
|
|
|
|
|
|
required_columns = ["Mean_Respiratory_Rate_BPM", "Number_of_Peaks", "Peak_Times_Seconds"]
|
|
|
for df, name in [(output_df, "Output"), (gt_df, "Ground Truth")]:
|
|
|
missing_cols = [col for col in required_columns if col not in df.columns]
|
|
|
if missing_cols:
|
|
|
print(f"Error: {name} CSV missing columns: {missing_cols}")
|
|
|
return result
|
|
|
|
|
|
|
|
|
pred_rate = output_df["Mean_Respiratory_Rate_BPM"].iloc[0]
|
|
|
pred_peaks = json.loads(output_df["Peak_Times_Seconds"].iloc[0])
|
|
|
pred_count = output_df["Number_of_Peaks"].iloc[0]
|
|
|
|
|
|
gt_rate = gt_df["Mean_Respiratory_Rate_BPM"].iloc[0]
|
|
|
gt_peaks = json.loads(gt_df["Peak_Times_Seconds"].iloc[0])
|
|
|
gt_count = gt_df["Number_of_Peaks"].iloc[0]
|
|
|
|
|
|
|
|
|
rate_mae = abs(pred_rate - gt_rate) if not np.isnan(pred_rate) and not np.isnan(gt_rate) else np.nan
|
|
|
rate_success = rate_mae <= 1.0 if not np.isnan(rate_mae) else False
|
|
|
|
|
|
if pred_peaks and gt_peaks:
|
|
|
peak_errors = [min(abs(p - gt) for gt in gt_peaks) for p in pred_peaks]
|
|
|
peak_mptd = sum(peak_errors) / len(peak_errors) if peak_errors else np.nan
|
|
|
peak_matching_count = sum(1 for err in peak_errors if err <= 0.1)
|
|
|
peak_matching_rate = peak_matching_count / len(pred_peaks) if pred_peaks else 0.0
|
|
|
else:
|
|
|
peak_mptd = np.nan
|
|
|
peak_matching_rate = 0.0
|
|
|
peak_success = (peak_mptd <= 0.1 and peak_matching_rate >= 0.8) if not np.isnan(peak_mptd) else False
|
|
|
|
|
|
peak_count_relative_error = (
|
|
|
abs(pred_count - gt_count) / gt_count if gt_count > 0 else np.nan
|
|
|
)
|
|
|
count_success = peak_count_relative_error <= 0.1 if not np.isnan(peak_count_relative_error) else False
|
|
|
|
|
|
success = rate_success and peak_success and count_success
|
|
|
result["Result"] = success
|
|
|
|
|
|
print("Evaluation Results:")
|
|
|
print(f"Rate MAE: {rate_mae:.2f} BPM (Success: {rate_success})")
|
|
|
print(f"Peak MPTD: {peak_mptd:.3f} seconds (Success: {peak_success})")
|
|
|
print(f"Peak Matching Rate: {peak_matching_rate:.2f}")
|
|
|
print(f"Peak Count Relative Error: {peak_count_relative_error:.2f} (Success: {count_success})")
|
|
|
print(f"Overall Success: {success}")
|
|
|
|
|
|
except Exception as e:
|
|
|
traceback.print_exc(file=log_output)
|
|
|
|
|
|
finally:
|
|
|
sys.stdout = sys_stdout
|
|
|
result["comments"] = log_output.getvalue()
|
|
|
return result
|
|
|
|
|
|
def main():
|
|
|
parser = argparse.ArgumentParser(description="Evaluate RSP metrics against ground truth")
|
|
|
parser.add_argument("--output", required=True, help="Path to output CSV file (rsp_metrics.csv)")
|
|
|
parser.add_argument("--groundtruth", required=True, help="Path to ground truth CSV file")
|
|
|
parser.add_argument("--result", default="results.jsonl", help="Path to JSONL result file")
|
|
|
args = parser.parse_args()
|
|
|
|
|
|
result_data = evaluate_rsp_metrics(args.output, args.groundtruth)
|
|
|
|
|
|
|
|
|
try:
|
|
|
with open(args.result, "a", encoding='utf-8') as f:
|
|
|
f.write(json.dumps(result_data, ensure_ascii=False, default=str) + "\n")
|
|
|
except Exception as e:
|
|
|
print(f"Failed to write result to {args.result}: {e}")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
main() |
