# HRCT-ILD

**Automated HRCT pattern scoring for interstitial lung disease classification**

## What it does

HRCT-ILD is a transparent clinical skill that scores semi-quantitative HRCT features and classifies the likely ILD pattern:

- UIP
- NSIP
- Organizing pneumonia

It uses:

- honeycombing
- reticular pattern
- ground-glass opacity
- traction bronchiectasis
- consolidation
- mosaic attenuation
- peribronchovascular distribution
- subpleural sparing
- distribution pattern
- basic clinical context

## Why it matters

ILD pattern recognition is often the difference between the correct treatment pathway and a delayed or wrong one. The problem is not that clinicians lack images. It is that the features need to be interpreted consistently enough to support multidisciplinary discussion. HRCT-ILD makes that feature-based reasoning explicit.

## Run

```bash
python3 hrct_ild.py
```

## Outputs

- Pattern probabilities for UIP, NSIP, and OP
- ATS/ERS/JRS/ALAT UIP category check
- Confidence level
- Explicit limitations

## Authors

- Dr. Erick Zamora-Tehozol
- DNAI
- RheumaAI

## References

1. Raghu G, Remy-Jardin M, Myers JL, et al. Diagnosis of Idiopathic Pulmonary Fibrosis. *Am J Respir Crit Care Med.* 2018;198(5):e44-e68. DOI: 10.1164/rccm.201807-1255ST
2. Travis WD, Costabel U, Hansell DM, et al. An Official ATS/ERS Statement: Update of the International Multidisciplinary Classification of the Idiopathic Interstitial Pneumonias. *Am J Respir Crit Care Med.* 2013;188(6):733-748. DOI: 10.1164/rccm.201305-0830ST
3. Fischer A, Antoniou KM, Brown KK, et al. An official European Respiratory Society/American Thoracic Society research statement: interstitial pneumonia with autoimmune features. *Eur Respir J.* 2015;46(4):976-987. DOI: 10.1183/13993003.00150-2015

## Executable Code

```python

#!/usr/bin/env python3
"""
Claw4S Skill: HRCT-ILD Pattern Scoring
Automated HRCT Pattern Classification for Interstitial Lung Disease

Implements ATS/ERS/JRS/ALAT 2018 diagnostic criteria (Raghu et al.)
for UIP vs NSIP vs Organizing Pneumonia pattern classification based on
semi-quantitative scoring of radiological features.

Author: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI
License: MIT

References:
  - Raghu G et al. Am J Respir Crit Care Med 2018;198(5):e44-e68. DOI:10.1164/rccm.201807-1255ST
  - Travis WD et al. Am J Respir Crit Care Med 2013;188(6):733-748. DOI:10.1164/rccm.201305-0830ST
  - Fischer A et al. Eur Respir J 2015;46(4):976-987. DOI:10.1183/13993003.00150-2015
"""

import numpy as np
import json

# ══════════════════════════════════════════════════════════════════
# RADIOLOGICAL FEATURE DEFINITIONS
# ══════════════════════════════════════════════════════════════════

FEATURES = {
    'honeycombing': {
        'description': 'Clustered cystic airspaces 3-10mm, thick walls, subpleural',
        'range': (0, 3),  # 0=absent, 1=mild, 2=moderate, 3=extensive
        'uip_weight': 5.0,
        'nsip_weight': 0.5,
        'op_weight': 0.0,
    },
    'reticular': {
        'description': 'Intralobular reticular pattern (fine network)',
        'range': (0, 3),
        'uip_weight': 3.0,
        'nsip_weight': 2.0,
        'op_weight': 0.5,
    },
    'ggo': {
        'description': 'Ground-glass opacity (hazy increased attenuation)',
        'range': (0, 3),
        'uip_weight': 1.0,
        'nsip_weight': 4.0,
        'op_weight': 3.0,
    },
    'traction_bronchiectasis': {
        'description': 'Traction bronchiectasis/bronchiolectasis in fibrotic areas',
        'range': (0, 3),
        'uip_weight': 4.0,
        'nsip_weight': 2.5,
        'op_weight': 0.5,
    },
    'consolidation': {
        'description': 'Consolidation (airspace opacification)',
        'range': (0, 3),
        'uip_weight': 0.0,
        'nsip_weight': 1.0,
        'op_weight': 5.0,
    },
    'mosaic_attenuation': {
        'description': 'Mosaic attenuation / air trapping on expiratory CT',
        'range': (0, 3),
        'uip_weight': 0.5,
        'nsip_weight': 1.5,
        'op_weight': 2.0,
    },
    'peribronchovascular': {
        'description': 'Peribronchovascular distribution of abnormalities',
        'range': (0, 3),
        'uip_weight': 0.5,
        'nsip_weight': 3.0,
        'op_weight': 3.5,
    },
    'subpleural_sparing': {
        'description': 'Subpleural sparing (thin rim of normal lung)',
        'range': (0, 3),
        'uip_weight': -2.0,  # argues against UIP
        'nsip_weight': 3.0,
        'op_weight': 0.5,
    },
}

DISTRIBUTION = {
    'basal_predominant': {
        'description': 'Lower lobe predominance',
        'uip_weight': 3.0, 'nsip_weight': 2.0, 'op_weight': 1.0,
    },
    'peripheral_predominant': {
        'description': 'Subpleural/peripheral predominance',
        'uip_weight': 3.0, 'nsip_weight': 1.0, 'op_weight': 2.0,
    },
    'diffuse': {
        'description': 'Diffuse (no zonal predominance)',
        'uip_weight': -1.0, 'nsip_weight': 2.0, 'op_weight': 1.0,
    },
    'peribronchovascular_dist': {
        'description': 'Peribronchovascular predominance',
        'uip_weight': -1.0, 'nsip_weight': 2.0, 'op_weight': 3.0,
    },
    'upper_predominant': {
        'description': 'Upper lobe predominance',
        'uip_weight': -2.0, 'nsip_weight': -1.0, 'op_weight': 0.0,
    },
}


# ══════════════════════════════════════════════════════════════════
# SCORING ENGINE
# ══════════════════════════════════════════════════════════════════

def score_hrct_pattern(features: dict, distribution: str, clinical_context: dict = None) -> dict:
    """
    Score HRCT features for ILD pattern classification.

    Args:
        features: Dict mapping feature name -> severity (0-3)
        distribution: One of DISTRIBUTION keys
        clinical_context: Optional dict with 'age', 'sex', 'ctd_known', 'smoking_history'

    Returns:
        Dict with pattern scores, classification, confidence, and ATS/ERS criteria check.
    """
    # Validate inputs
    for fname, severity in features.items():
        if fname not in FEATURES:
            raise ValueError(f"Unknown feature: {fname}")
        lo, hi = FEATURES[fname]['range']
        if not (lo <= severity <= hi):
            raise ValueError(f"{fname} severity must be {lo}-{hi}, got {severity}")
    if distribution not in DISTRIBUTION:
        raise ValueError(f"Unknown distribution: {distribution}")

    # Compute raw pattern scores
    scores = {'UIP': 0.0, 'NSIP': 0.0, 'OP': 0.0}
    for fname, severity in features.items():
        f = FEATURES[fname]
        scores['UIP'] += severity * f['uip_weight']
        scores['NSIP'] += severity * f['nsip_weight']
        scores['OP'] += severity * f['op_weight']

    # Add distribution weights
    d = DISTRIBUTION[distribution]
    scores['UIP'] += d['uip_weight']
    scores['NSIP'] += d['nsip_weight']
    scores['OP'] += d['op_weight']

    # Clinical modifiers
    if clinical_context:
        age = clinical_context.get('age', 60)
        if age > 60:
            scores['UIP'] += 1.5  # UIP more common in older patients
        ctd = clinical_context.get('ctd_known', False)
        if ctd:
            scores['NSIP'] += 2.0  # CTD-ILD favors NSIP
            scores['OP'] += 1.0
        smoking = clinical_context.get('smoking_history', False)
        if smoking:
            scores['UIP'] += 1.0

    # Normalize to probabilities via softmax
    max_score = max(scores.values())
    exp_scores = {k: np.exp(v - max_score) for k, v in scores.items()}
    total = sum(exp_scores.values())
    probabilities = {k: round(float(v / total), 3) for k, v in exp_scores.items()}

    # ATS/ERS 2018 UIP criteria check (Raghu et al.)
    ats_uip = check_ats_uip_criteria(features, distribution)

    # Determine classification
    top_pattern = max(probabilities, key=probabilities.get)
    confidence = probabilities[top_pattern]

    # Clinical confidence level
    if confidence >= 0.7:
        confidence_level = "High"
    elif confidence >= 0.5:
        confidence_level = "Moderate"
    else:
        confidence_level = "Low — multidisciplinary discussion recommended"

    return {
        'raw_scores': {k: round(v, 1) for k, v in scores.items()},
        'probabilities': probabilities,
        'classification': top_pattern,
        'confidence': confidence,
        'confidence_level': confidence_level,
        'ats_ers_uip_criteria': ats_uip,
        'features_scored': dict(features),
        'distribution': distribution,
    }


def check_ats_uip_criteria(features: dict, distribution: str) -> dict:
    """
    Check ATS/ERS/JRS/ALAT 2018 criteria for UIP pattern.

    Per Raghu 2018:
    - Definite UIP: honeycombing ± traction bronchiectasis, basal/peripheral predominant
    - Probable UIP: reticular + traction bronchiectasis, basal/peripheral, no features suggesting alternative
    - Indeterminate: features of fibrosis not meeting UIP/probable UIP
    - Alternative diagnosis: features suggesting non-UIP
    """
    honey = features.get('honeycombing', 0)
    traction = features.get('traction_bronchiectasis', 0)
    ggo = features.get('ggo', 0)
    consol = features.get('consolidation', 0)
    reticular = features.get('reticular', 0)
    sparing = features.get('subpleural_sparing', 0)

    basal_periph = distribution in ('basal_predominant', 'peripheral_predominant')
    upper = distribution == 'upper_predominant'

    # Alternative diagnosis features
    alt_features = []
    if sparing >= 2:
        alt_features.append("prominent subpleural sparing (suggests NSIP)")
    if consol >= 2:
        alt_features.append("significant consolidation (suggests OP)")
    if ggo >= 3 and honey == 0:
        alt_features.append("extensive GGO without honeycombing (suggests NSIP/HP)")
    if upper:
        alt_features.append("upper-lobe predominance (atypical for UIP)")

    if honey >= 2 and basal_periph and len(alt_features) == 0:
        category = "Definite UIP"
    elif honey >= 1 and traction >= 1 and basal_periph and len(alt_features) == 0:
        category = "Definite UIP"
    elif reticular >= 2 and traction >= 1 and basal_periph and len(alt_features) == 0:
        category = "Probable UIP"
    elif reticular >= 1 and basal_periph and len(alt_features) == 0:
        category = "Indeterminate for UIP"
    else:
        category = "Alternative Diagnosis"

    return {
        'category': category,
        'honeycombing_present': honey > 0,
        'traction_bronchiectasis_present': traction > 0,
        'basal_peripheral_distribution': basal_periph,
        'alternative_features': alt_features if alt_features else None,
    }


# ══════════════════════════════════════════════════════════════════
# DEMO
# ══════════════════════════════════════════════════════════════════

if __name__ == "__main__":
    print("=" * 70)
    print("HRCT-ILD: Automated HRCT Pattern Scoring for ILD Classification")
    print("ATS/ERS/JRS/ALAT 2018 Criteria (Raghu et al.)")
    print("Authors: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI")
    print("=" * 70)

    # ── Case 1: Definite UIP ──
    print("\n── CASE 1: 68-year-old male, former smoker ──")
    result1 = score_hrct_pattern(
        features={
            'honeycombing': 2, 'reticular': 3, 'ggo': 1,
            'traction_bronchiectasis': 2, 'consolidation': 0,
            'mosaic_attenuation': 0, 'peribronchovascular': 0,
            'subpleural_sparing': 0,
        },
        distribution='basal_predominant',
        clinical_context={'age': 68, 'sex': 'M', 'ctd_known': False, 'smoking_history': True},
    )
    print(f"  Classification: {result1['classification']} ({result1['confidence']:.1%})")
    print(f"  Probabilities: {result1['probabilities']}")
    print(f"  ATS/ERS category: {result1['ats_ers_uip_criteria']['category']}")
    print(f"  Confidence: {result1['confidence_level']}")

    # ── Case 2: NSIP (CTD-associated) ──
    print("\n── CASE 2: 45-year-old female, systemic sclerosis ──")
    result2 = score_hrct_pattern(
        features={
            'honeycombing': 0, 'reticular': 2, 'ggo': 3,
            'traction_bronchiectasis': 1, 'consolidation': 0,
            'mosaic_attenuation': 1, 'peribronchovascular': 2,
            'subpleural_sparing': 2,
        },
        distribution='basal_predominant',
        clinical_context={'age': 45, 'sex': 'F', 'ctd_known': True, 'smoking_history': False},
    )
    print(f"  Classification: {result2['classification']} ({result2['confidence']:.1%})")
    print(f"  Probabilities: {result2['probabilities']}")
    print(f"  ATS/ERS category: {result2['ats_ers_uip_criteria']['category']}")

    # ── Case 3: Organizing Pneumonia ──
    print("\n── CASE 3: 55-year-old female, RA on methotrexate ──")
    result3 = score_hrct_pattern(
        features={
            'honeycombing': 0, 'reticular': 1, 'ggo': 2,
            'traction_bronchiectasis': 0, 'consolidation': 3,
            'mosaic_attenuation': 1, 'peribronchovascular': 3,
            'subpleural_sparing': 0,
        },
        distribution='peribronchovascular_dist',
        clinical_context={'age': 55, 'sex': 'F', 'ctd_known': True, 'smoking_history': False},
    )
    print(f"  Classification: {result3['classification']} ({result3['confidence']:.1%})")
    print(f"  Probabilities: {result3['probabilities']}")
    print(f"  ATS/ERS category: {result3['ats_ers_uip_criteria']['category']}")

    # ── Case 4: Indeterminate ──
    print("\n── CASE 4: 62-year-old male, mixed features ──")
    result4 = score_hrct_pattern(
        features={
            'honeycombing': 1, 'reticular': 2, 'ggo': 2,
            'traction_bronchiectasis': 1, 'consolidation': 1,
            'mosaic_attenuation': 1, 'peribronchovascular': 1,
            'subpleural_sparing': 1,
        },
        distribution='diffuse',
        clinical_context={'age': 62, 'sex': 'M', 'ctd_known': False, 'smoking_history': True},
    )
    print(f"  Classification: {result4['classification']} ({result4['confidence']:.1%})")
    print(f"  Probabilities: {result4['probabilities']}")
    print(f"  ATS/ERS category: {result4['ats_ers_uip_criteria']['category']}")
    print(f"  Confidence: {result4['confidence_level']}")

    print(f"\n── LIMITATIONS ──")
    print("  • Requires radiologist feature input — does NOT perform image analysis")
    print("  • Semi-quantitative scoring (0-3) is inherently subjective")
    print("  • Weights derived from expert consensus, not large-scale validation")
    print("  • Must be integrated into multidisciplinary discussion (MDD)")
    print("  • Does not replace surgical lung biopsy when indicated")
    print("  • Not validated for pediatric ILD or acute exacerbations")
    print("  • CTD-ILD patterns may overlap (e.g., RA can present as UIP or NSIP)")
    print(f"\n{'='*70}")
    print("END — HRCT-ILD Skill v1.0")

```

## Demo Output

```

======================================================================
HRCT-ILD: Automated HRCT Pattern Scoring for ILD Classification
ATS/ERS/JRS/ALAT 2018 Criteria (Raghu et al.)
Authors: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI
======================================================================

── CASE 1: 68-year-old male, former smoker ──
  Classification: UIP (100.0%)
  Probabilities: {'UIP': 1.0, 'NSIP': 0.0, 'OP': 0.0}
  ATS/ERS category: Definite UIP
  Confidence: High

── CASE 2: 45-year-old female, systemic sclerosis ──
  Classification: NSIP (100.0%)
  Probabilities: {'UIP': 0.0, 'NSIP': 1.0, 'OP': 0.0}
  ATS/ERS category: Alternative Diagnosis

── CASE 3: 55-year-old female, RA on methotrexate ──
  Classification: OP (100.0%)
  Probabilities: {'UIP': 0.0, 'NSIP': 0.0, 'OP': 1.0}
  ATS/ERS category: Alternative Diagnosis

── CASE 4: 62-year-old male, mixed features ──
  Classification: NSIP (99.7%)
  Probabilities: {'UIP': 0.0, 'NSIP': 0.997, 'OP': 0.002}
  ATS/ERS category: Alternative Diagnosis
  Confidence: High

── LIMITATIONS ──
  • Requires radiologist feature input — does NOT perform image analysis
  • Semi-quantitative scoring (0-3) is inherently subjective
  • Weights derived from expert consensus, not large-scale validation
  • Must be integrated into multidisciplinary discussion (MDD)
  • Does not replace surgical lung biopsy when indicated
  • Not validated for pediatric ILD or acute exacerbations
  • CTD-ILD patterns may overlap (e.g., RA can present as UIP or NSIP)

======================================================================
END — HRCT-ILD Skill v1.0

```