|
|
|
|
|
""" |
|
|
saggital_ICC.py |
|
|
Recompute ICC analysis for sagittal measurements only using the 2 CSV files. |
|
|
Each column represents data from one rater, comparing all 5 columns for ICC. |
|
|
""" |
|
|
|
|
|
import pandas as pd |
|
|
import numpy as np |
|
|
import matplotlib.pyplot as plt |
|
|
from scipy import stats |
|
|
from scipy.stats import f |
|
|
import argparse |
|
|
import sys |
|
|
from pathlib import Path |
|
|
|
|
|
ID_LIKE = {"case", "case_id", "id", "subject", "subject_id", "uid", "study", "study_id"} |
|
|
|
|
|
def detect_rater_columns(df: pd.DataFrame, min_unique: int = 3): |
|
|
"""Detect rater columns.""" |
|
|
rater_like = [c for c in df.columns if str(c).strip().lower().startswith("rater")] |
|
|
if len(rater_like) >= 2: |
|
|
return rater_like |
|
|
num_cols = df.select_dtypes(include=[np.number]).columns.tolist() |
|
|
keep = [] |
|
|
for c in num_cols: |
|
|
cl = str(c).strip().lower() |
|
|
if cl in ID_LIKE: |
|
|
continue |
|
|
if df[c].nunique(dropna=True) >= min_unique: |
|
|
keep.append(c) |
|
|
return keep |
|
|
|
|
|
def icc2k_absolute(y: np.ndarray): |
|
|
""" |
|
|
Compute ICC(2,k): two-way random-effects, absolute-agreement, average of k raters. |
|
|
Returns ICC(2,k) and mean-square terms. |
|
|
""" |
|
|
y = np.asarray(y, float) |
|
|
n, k = y.shape |
|
|
mean_targets = y.mean(axis=1, keepdims=True) |
|
|
mean_raters = y.mean(axis=0, keepdims=True) |
|
|
grand_mean = y.mean() |
|
|
|
|
|
SSR = k * np.sum((mean_targets - grand_mean)**2) |
|
|
SSC = n * np.sum((mean_raters - grand_mean)**2) |
|
|
SSE = np.sum((y - mean_targets - mean_raters + grand_mean)**2) |
|
|
|
|
|
dfR, dfC = n - 1, k - 1 |
|
|
dfE = (n - 1) * (k - 1) |
|
|
|
|
|
MSR = SSR / dfR if dfR > 0 else np.nan |
|
|
MSC = SSC / dfC if dfC > 0 else np.nan |
|
|
MSE = SSE / dfE if dfE > 0 else np.nan |
|
|
|
|
|
numerator = MSR - MSE |
|
|
denominator = MSR + (MSC - MSE) / n |
|
|
icc2k = numerator / denominator if denominator != 0 else np.nan |
|
|
return icc2k, MSR, MSC, MSE |
|
|
|
|
|
def bootstrap_icc2k(y, n_boot=5000, seed=42): |
|
|
"""Bootstrap ICC(2,k) confidence intervals.""" |
|
|
rng = np.random.default_rng(seed) |
|
|
n, _ = y.shape |
|
|
boots = [] |
|
|
for _ in range(n_boot): |
|
|
idx = rng.integers(0, n, size=n) |
|
|
icc, _, _, _ = icc2k_absolute(y[idx, :]) |
|
|
boots.append(icc) |
|
|
boots = np.asarray(boots) |
|
|
lo, hi = np.nanpercentile(boots, [2.5, 97.5]) |
|
|
return float(np.nanmean(boots)), float(lo), float(hi), boots |
|
|
|
|
|
def format_pm(mean, sd, decimals=1): |
|
|
"""Format mean ± SD.""" |
|
|
if np.isnan(mean) or np.isnan(sd): |
|
|
return "NA" |
|
|
f = f"{{:.{decimals}f}} ± {{:.{decimals}f}}" |
|
|
return f.format(mean, sd) |
|
|
|
|
|
def detect_cobb_series(df: pd.DataFrame) -> pd.Series: |
|
|
"""Detect Cobb angle column in test data.""" |
|
|
|
|
|
cobb_cols = [c for c in df.columns if "cobb" in str(c).lower()] |
|
|
if cobb_cols: |
|
|
s = pd.to_numeric(df[cobb_cols[0]], errors="coerce") |
|
|
return s |
|
|
|
|
|
num_cols = df.select_dtypes(include=[np.number]).columns.tolist() |
|
|
if not num_cols: |
|
|
raise ValueError("No numeric columns found for Cobb angles.") |
|
|
return df[num_cols[0]] |
|
|
|
|
|
def fmt(x, dec=1): |
|
|
return f"{x:.{dec}f}" |
|
|
|
|
|
def create_test_cobb_summary(csv_path, outdir=".", decimals=1): |
|
|
"""Create summary statistics for test dataset with single-observer Cobb angles.""" |
|
|
csv_path = Path(csv_path) |
|
|
outdir = Path(outdir) |
|
|
outdir.mkdir(parents=True, exist_ok=True) |
|
|
|
|
|
if not csv_path.exists(): |
|
|
print(f"[ERROR] CSV not found: {csv_path}") |
|
|
return |
|
|
|
|
|
df = pd.read_csv(csv_path) |
|
|
try: |
|
|
s = detect_cobb_series(df) |
|
|
except Exception as e: |
|
|
print(f"[ERROR] {e}") |
|
|
print(f"Columns seen: {list(df.columns)}") |
|
|
return |
|
|
|
|
|
x = pd.to_numeric(s, errors="coerce").dropna().to_numpy() |
|
|
n = x.size |
|
|
if n == 0: |
|
|
print("[ERROR] No valid numeric Cobb values found.") |
|
|
return |
|
|
|
|
|
mean = float(np.mean(x)) |
|
|
sd = float(np.std(x, ddof=1)) if n > 1 else float("nan") |
|
|
median = float(np.median(x)) |
|
|
q1, q3 = [float(np.percentile(x, p)) for p in (25, 75)] |
|
|
iqr = q3 - q1 |
|
|
xmin = float(np.min(x)) |
|
|
xmax = float(np.max(x)) |
|
|
|
|
|
print("\n=== Single-Observer Thoracic Cobb Summary (Test Set) ===") |
|
|
print(f"n = {n}") |
|
|
print(f"Mean ± SD: {fmt(mean, decimals)} ± {fmt(sd, decimals)} deg") |
|
|
print(f"Median [IQR]: {fmt(median, decimals)} [{fmt(q1, decimals)}–{fmt(q3, decimals)}] deg") |
|
|
print(f"Range: {fmt(xmin, decimals)}–{fmt(xmax, decimals)} deg") |
|
|
print("=========================================================") |
|
|
|
|
|
out_csv = outdir / "test_cobb_summary.csv" |
|
|
pd.DataFrame([{ |
|
|
"n": n, |
|
|
"mean": mean, |
|
|
"sd": sd, |
|
|
"median": median, |
|
|
"q1": q1, |
|
|
"q3": q3, |
|
|
"iqr": iqr, |
|
|
"min": xmin, |
|
|
"max": xmax |
|
|
}]).to_csv(out_csv, index=False) |
|
|
|
|
|
print(f"[OK] Saved: {out_csv}") |
|
|
return mean, sd, median, q1, q3, iqr, xmin, xmax, n |
|
|
|
|
|
def calculate_icc_2_1(data): |
|
|
"""Calculate ICC(2,1).""" |
|
|
n_subjects, n_raters = data.shape |
|
|
|
|
|
subject_means = np.mean(data, axis=1) |
|
|
rater_means = np.mean(data, axis=0) |
|
|
grand_mean = np.mean(data) |
|
|
|
|
|
SS_between_subjects = n_raters * np.sum((subject_means - grand_mean) ** 2) |
|
|
SS_between_raters = n_subjects * np.sum((rater_means - grand_mean) ** 2) |
|
|
SS_error = 0 |
|
|
for i in range(n_subjects): |
|
|
for j in range(n_raters): |
|
|
SS_error += (data[i, j] - subject_means[i] - rater_means[j] + grand_mean) ** 2 |
|
|
|
|
|
MS_between_subjects = SS_between_subjects / (n_subjects - 1) |
|
|
MS_between_raters = SS_between_raters / (n_raters - 1) |
|
|
MS_error = SS_error / ((n_subjects - 1) * (n_raters - 1)) |
|
|
|
|
|
icc_numerator = MS_between_subjects - MS_error |
|
|
icc_denominator = MS_between_subjects + (n_raters - 1) * MS_error |
|
|
icc_value = icc_numerator / icc_denominator |
|
|
|
|
|
f_stat = MS_between_subjects / MS_error |
|
|
df1 = n_subjects - 1 |
|
|
df2 = (n_subjects - 1) * (n_raters - 1) |
|
|
p_value = 1 - f.cdf(f_stat, df1, df2) |
|
|
|
|
|
alpha = 0.05 |
|
|
f_lower = f_stat / f.ppf(1 - alpha/2, df1, df2) |
|
|
f_upper = f_stat * f.ppf(1 - alpha/2, df1, df2) |
|
|
|
|
|
ci_lower = max(0, (f_lower - 1) / (f_lower + n_raters - 1)) |
|
|
ci_upper = min(1, (f_upper - 1) / (f_upper + n_raters - 1)) |
|
|
|
|
|
return icc_value, f_stat, p_value, (ci_lower, ci_upper) |
|
|
|
|
|
def create_comprehensive_summary(csv_path, outdir=".", decimals=1, n_boot=5000): |
|
|
"""Create summary statistics including bootstrap ICC.""" |
|
|
csv_path = Path(csv_path) |
|
|
outdir = Path(outdir) |
|
|
outdir.mkdir(parents=True, exist_ok=True) |
|
|
|
|
|
if not csv_path.exists(): |
|
|
print(f"[ERROR] CSV not found: {csv_path}") |
|
|
return |
|
|
|
|
|
df = pd.read_csv(csv_path, sep='\t', header=None) |
|
|
|
|
|
raters = list(range(df.shape[1])) |
|
|
y = df.to_numpy(float) |
|
|
n, k = y.shape |
|
|
|
|
|
rater_means = np.nanmean(y, axis=0) |
|
|
rater_sds = np.nanstd(y, axis=0, ddof=1) |
|
|
|
|
|
per_case_sd = np.nanstd(y, axis=1, ddof=1) |
|
|
across_mean = float(np.nanmean(per_case_sd)) |
|
|
across_sd = float(np.nanstd(per_case_sd, ddof=1)) |
|
|
|
|
|
grand_mean = float(np.nanmean(y)) |
|
|
|
|
|
icc2k, MSR, MSC, MSE = icc2k_absolute(y) |
|
|
_, lo, hi, boots = bootstrap_icc2k(y, n_boot=n_boot) |
|
|
|
|
|
print("\n=== Five-Observer Thoracic Cobb Summary (Development Set) ===") |
|
|
print(f"Detected raters (k={k}): {raters}") |
|
|
for i, (m, s) in enumerate(zip(rater_means, rater_sds)): |
|
|
print(f"Rater {i+1:>8d}: {m:.{decimals}f} ± {s:.{decimals}f} deg") |
|
|
print(f"Across-rater SD (per-case): mean ± SD = {across_mean:.{decimals}f} ± {across_sd:.{decimals}f} deg") |
|
|
print(f"Grand mean across all ratings: {grand_mean:.{decimals}f} deg") |
|
|
print(f"ICC(2,k) absolute agreement (bootstrap 95% CI): {icc2k:.3f} [{lo:.3f}, {hi:.3f}]") |
|
|
print("==============================================================") |
|
|
|
|
|
rows = [] |
|
|
for i, (m, s) in enumerate(zip(rater_means, rater_sds)): |
|
|
rows.append({"measure": "rater_mean_sd", "rater": f"Rater_{i+1}", "mean": m, "sd": s}) |
|
|
rows.append({"measure": "across_rater_sd_mean", "rater": "NA", "mean": across_mean, "sd": across_sd}) |
|
|
rows.append({"measure": "grand_mean", "rater": "NA", "mean": grand_mean, "sd": np.nan}) |
|
|
rows.append({"measure": "icc2k", "rater": "NA", "mean": icc2k, "sd": np.nan}) |
|
|
rows.append({"measure": "icc2k_ci_lo", "rater": "NA", "mean": lo, "sd": np.nan}) |
|
|
rows.append({"measure": "icc2k_ci_hi", "rater": "NA", "mean": hi, "sd": np.nan}) |
|
|
pd.DataFrame(rows).to_csv(outdir / "dev_cobb_summary.csv", index=False) |
|
|
|
|
|
print(f"[OK] Saved summaries in {outdir.resolve()}") |
|
|
return icc2k, lo, hi |
|
|
|
|
|
def create_sagittal_icc_plot(): |
|
|
"""Create ICC plot for sagittal measurements only""" |
|
|
|
|
|
csv_files = { |
|
|
'../cobb_angles/dev_cobb.csv': 'Sagittal Thoracic' |
|
|
} |
|
|
|
|
|
results = {} |
|
|
for filename, display_name in csv_files.items(): |
|
|
try: |
|
|
df = pd.read_csv(filename, sep='\t', header=None) |
|
|
data = df.values |
|
|
print(f"\n{display_name} Data Shape: {data.shape}") |
|
|
print(f"Data preview:\n{data[:5]}") |
|
|
|
|
|
icc_value, f_stat, p_value, ci = calculate_icc_2_1(data) |
|
|
|
|
|
results[display_name] = { |
|
|
'icc': icc_value, |
|
|
'f_stat': f_stat, |
|
|
'p_value': p_value, |
|
|
'ci_lower': ci[0], |
|
|
'ci_upper': ci[1], |
|
|
'n_subjects': data.shape[0], |
|
|
'n_raters': data.shape[1] |
|
|
} |
|
|
|
|
|
print(f"{display_name}: ICC = {icc_value:.4f}, CI = [{ci[0]:.3f}, {ci[1]:.3f}]") |
|
|
print(f"F-statistic = {f_stat:.4f}, p-value = {p_value:.4f}") |
|
|
|
|
|
except Exception as e: |
|
|
print(f"Error processing {filename}: {e}") |
|
|
continue |
|
|
|
|
|
if not results: |
|
|
print("No data processed successfully.") |
|
|
return |
|
|
|
|
|
fig1, ax1 = plt.subplots(1, 1, figsize=(4, 8)) |
|
|
|
|
|
names = list(results.keys()) |
|
|
icc_values = [results[name]['icc'] for name in names] |
|
|
ci_lowers = [results[name]['ci_lower'] for name in names] |
|
|
ci_uppers = [results[name]['ci_upper'] for name in names] |
|
|
|
|
|
colors = ['#2E86AB', '#A23B72'] |
|
|
|
|
|
bars = ax1.bar(names, icc_values, color=colors, alpha=0.8, width=0.3, |
|
|
edgecolor='black', linewidth=1) |
|
|
|
|
|
ax1.errorbar(names, icc_values, |
|
|
yerr=[np.array(icc_values) - np.array(ci_lowers), |
|
|
np.array(ci_uppers) - np.array(icc_values)], |
|
|
fmt='none', color='red', capsize=5, capthick=2) |
|
|
|
|
|
for i, (bar, value) in enumerate(zip(bars, icc_values)): |
|
|
ax1.text(bar.get_x() + bar.get_width()/2, value + 0.02, |
|
|
f'{value:.3f}', ha='center', va='bottom', |
|
|
fontweight='bold', fontsize=12) |
|
|
|
|
|
ax1.set_ylabel('ICC Value', fontsize=12, fontweight='bold') |
|
|
ax1.set_title('Intraclass Correlation Coefficients\nSagittal Thoracic Measurements', |
|
|
fontsize=14, fontweight='bold') |
|
|
ax1.set_ylim(0, 1.1) |
|
|
ax1.grid(True, alpha=0.3, linestyle='--') |
|
|
ax1.set_axisbelow(True) |
|
|
|
|
|
ax1.tick_params(axis='x', rotation=0) |
|
|
|
|
|
plt.tight_layout() |
|
|
plt.show() |
|
|
|
|
|
fig2, ax2 = plt.subplots(1, 1, figsize=(4, 8)) |
|
|
ax2.axis('off') |
|
|
|
|
|
table_data = [] |
|
|
for name in names: |
|
|
result = results[name] |
|
|
table_data.append([ |
|
|
name, |
|
|
f"{result['icc']:.4f}", |
|
|
f"[{result['ci_lower']:.3f}, {result['ci_upper']:.3f}]", |
|
|
f"{result['p_value']:.4f}", |
|
|
f"{result['n_subjects']}x{result['n_raters']}" |
|
|
]) |
|
|
|
|
|
table = ax2.table(cellText=table_data, |
|
|
colLabels=['Measurement Type', 'ICC(2,1)', '95% CI', 'p-value', 'Dimensions'], |
|
|
cellLoc='center', |
|
|
loc='center', |
|
|
bbox=[0, 0, 1, 1]) |
|
|
|
|
|
table.auto_set_font_size(False) |
|
|
table.set_fontsize(10) |
|
|
table.scale(1, 2) |
|
|
|
|
|
for i in range(len(table_data[0])): |
|
|
table[(0, i)].set_facecolor('#4CAF50') |
|
|
table[(0, i)].set_text_props(weight='bold', color='white') |
|
|
|
|
|
for i in range(1, len(table_data) + 1): |
|
|
for j in range(len(table_data[0])): |
|
|
table[(i, j)].set_facecolor('#F5F5F5' if i % 2 == 0 else 'white') |
|
|
|
|
|
ax2.set_title('ICC Analysis Results - Sagittal Thoracic', fontsize=14, fontweight='bold', pad=20) |
|
|
|
|
|
plt.tight_layout() |
|
|
plt.show() |
|
|
|
|
|
if results: |
|
|
fig3, ax3 = plt.subplots(1, 1, figsize=(6, 6)) |
|
|
|
|
|
first_name = list(results.keys())[0] |
|
|
|
|
|
for filename, display_name in csv_files.items(): |
|
|
if display_name == first_name: |
|
|
df = pd.read_csv(filename, sep='\t', header=None) |
|
|
data = df.values |
|
|
break |
|
|
|
|
|
if data.shape[1] >= 2: |
|
|
|
|
|
means = np.mean(data, axis=1) |
|
|
|
|
|
differences = [] |
|
|
for i in range(data.shape[0]): |
|
|
subject_ratings = data[i, :] |
|
|
subject_mean = np.mean(subject_ratings) |
|
|
|
|
|
mean_abs_diff = np.mean(np.abs(subject_ratings - subject_mean)) |
|
|
differences.append(mean_abs_diff) |
|
|
differences = np.array(differences) |
|
|
|
|
|
mean_diff = np.mean(differences) |
|
|
std_diff = np.std(differences) |
|
|
upper_limit = mean_diff + 1.96 * std_diff |
|
|
lower_limit = mean_diff - 1.96 * std_diff |
|
|
|
|
|
ax3.scatter(means, differences, alpha=0.7, s=50, color='#2E86AB') |
|
|
ax3.axhline(y=mean_diff, color='red', linestyle='-', linewidth=2) |
|
|
ax3.axhline(y=upper_limit, color='red', linestyle='--', linewidth=1) |
|
|
ax3.axhline(y=lower_limit, color='red', linestyle='--', linewidth=1) |
|
|
|
|
|
ax3.set_xlabel('Mean Thoracic Cobb Angle of All Five Raters (deg)', fontsize=12, fontweight='bold') |
|
|
ax3.set_ylabel('Mean Absolute Difference from Average', fontsize=12, fontweight='bold') |
|
|
ax3.set_title('Inter-Rater Variability Plot\nSagittal Thoracic', fontsize=14, fontweight='bold') |
|
|
ax3.grid(True, alpha=0.3, linestyle='--') |
|
|
|
|
|
ax3.text(0.05, 0.95, f'Limits of Agreement:\n{lower_limit:.2f} to {upper_limit:.2f}', |
|
|
transform=ax3.transAxes, fontsize=10, verticalalignment='top', |
|
|
bbox=dict(boxstyle='round', facecolor='wheat', alpha=0.8)) |
|
|
|
|
|
plt.tight_layout() |
|
|
plt.savefig('../ICC_results/sagittal_inter_rater_variability.png', dpi=300, bbox_inches='tight', |
|
|
facecolor='white', edgecolor='none') |
|
|
plt.show() |
|
|
|
|
|
bland_altman_values = {} |
|
|
if results: |
|
|
first_name = list(results.keys())[0] |
|
|
for filename, display_name in csv_files.items(): |
|
|
if display_name == first_name: |
|
|
df = pd.read_csv(filename, sep='\t', header=None) |
|
|
data = df.values |
|
|
break |
|
|
|
|
|
if data.shape[1] >= 2: |
|
|
|
|
|
means = np.mean(data, axis=1) |
|
|
|
|
|
differences = [] |
|
|
for i in range(data.shape[0]): |
|
|
subject_ratings = data[i, :] |
|
|
subject_mean = np.mean(subject_ratings) |
|
|
|
|
|
mean_abs_diff = np.mean(np.abs(subject_ratings - subject_mean)) |
|
|
differences.append(mean_abs_diff) |
|
|
differences = np.array(differences) |
|
|
|
|
|
mean_diff = np.mean(differences) |
|
|
std_diff = np.std(differences) |
|
|
upper_limit = mean_diff + 1.96 * std_diff |
|
|
lower_limit = mean_diff - 1.96 * std_diff |
|
|
|
|
|
bland_altman_values = { |
|
|
'Mean_Difference': round(mean_diff, 2), |
|
|
'Upper_Limit': round(upper_limit, 2), |
|
|
'Lower_Limit': round(lower_limit, 2), |
|
|
'Std_Difference': round(std_diff, 2) |
|
|
} |
|
|
|
|
|
results_df = pd.DataFrame([ |
|
|
{ |
|
|
'Measurement_Type': name, |
|
|
'ICC_2_1': round(results[name]['icc'], 2), |
|
|
'CI_Lower': round(results[name]['ci_lower'], 2), |
|
|
'CI_Upper': round(results[name]['ci_upper'], 2), |
|
|
'F_Statistic': round(results[name]['f_stat'], 2), |
|
|
'P_Value': round(results[name]['p_value'], 2), |
|
|
'N_Subjects': results[name]['n_subjects'], |
|
|
'N_Raters': results[name]['n_raters'], |
|
|
'Bland_Altman_Mean_Diff': bland_altman_values.get('Mean_Difference', ''), |
|
|
'Bland_Altman_Upper_Limit': bland_altman_values.get('Upper_Limit', ''), |
|
|
'Bland_Altman_Lower_Limit': bland_altman_values.get('Lower_Limit', ''), |
|
|
'Bland_Altman_Std_Diff': bland_altman_values.get('Std_Difference', '') |
|
|
} |
|
|
for name in names |
|
|
]) |
|
|
|
|
|
results_df.to_csv('../ICC_results/sagittal_icc_results.csv', index=False) |
|
|
print(f"\nResults saved to '../ICC_results/sagittal_icc_results.csv'") |
|
|
print(f"Inter-Rater Variability Plot saved as '../ICC_results/sagittal_inter_rater_variability.png'") |
|
|
|
|
|
print(f"\n=== SAGITTAL THORACIC ICC ANALYSIS SUMMARY ===") |
|
|
for name in names: |
|
|
result = results[name] |
|
|
icc = result['icc'] |
|
|
ci_lower = result['ci_lower'] |
|
|
ci_upper = result['ci_upper'] |
|
|
|
|
|
print(f"{name}: ICC = {icc:.4f}") |
|
|
print(f" 95% CI: [{ci_lower:.3f}, {ci_upper:.3f}]") |
|
|
|
|
|
print("\n" + "="*60) |
|
|
print("SUMMARY WITH BOOTSTRAP ICC(2,k)") |
|
|
print("="*60) |
|
|
create_comprehensive_summary( |
|
|
csv_path='../cobb_angles/dev_cobb.csv', |
|
|
outdir='../ICC_results', |
|
|
decimals=1, |
|
|
n_boot=5000 |
|
|
) |
|
|
|
|
|
print("\n" + "="*60) |
|
|
print("TEST DATASET ANALYSIS (SINGLE-OBSERVER COBB ANGLES)") |
|
|
print("="*60) |
|
|
create_test_cobb_summary( |
|
|
csv_path='../cobb_angles/test_cobb.csv', |
|
|
outdir='../ICC_results', |
|
|
decimals=1 |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
create_sagittal_icc_plot() |
|
|
|