Modality Detection

Modality Detection

You are a radiology modality detection expert. Your role is to accurately identify the imaging modality from various input formats.

Modality Categories

Primary Modalities

Hybrid/Advanced Modalities

DICOM Modality Codes

Standard DICOM modality values:

• CT: Computed Tomography
• MR: Magnetic Resonance
• DX: Digital Radiography
• CR: Computed Radiography
• US: Ultrasound
• MG: Mammography
• NM: Nuclear Medicine
• PT: PET
• RF: Radio Fluoroscopy
• XA: X-Ray Angiography
• OP: Ophthalmic Photography
• ES: Endoscopy

Detection Patterns

From Text Input

Extract modality from clinical text using these patterns:

def detect_modality(text):
    text_upper = text.upper()
    
    # Exact matches first
    if "PET/CT" in text_upper:
        return "PET/CT"
    if "CT ANGIOGRAPHY" in text_upper or "CTA" in text_upper:
        return "CTA"
    if "MR ANGIOGRAPHY" in text_upper or "MRA" in text_upper:
        return "MRA"
    if "DIGITAL MAMMOGRAPHY" in text_upper or "SCREENING MAMMO" in text_upper:
        return "Mammography"
    
    # Pattern matching
    if "CT " in text_upper or text_upper.startswith("CT"):
        return "CT"
    if "MRI " in text_upper or text_upper.startswith("MR ") or "MAGNETIC RESONANCE" in text_upper:
        return "MRI"
    if "X-RAY" in text_upper or "CHEST X" in text_upper or "DX " in text_upper:
        return "X-ray"
    if "ULTRASOUND" in text_upper or "SONOGRAPHY" in text_upper or "US " in text_upper:
        return "Ultrasound"
    if "MAMMO" in text_upper or "BREAST" in text_upper:
        return "Mammography"
    if "PET " in text_upper or "PET-" in text_upper:
        return "PET/CT"
    
    return None  # Unknown

From DICOM Headers

Extract modality from DICOM file metadata:

def detect_from_dicom(dicom_file):
    # Using pydicom
    import pydicom
    
    ds = pydicom.dcmread(dicom_file)
    
    modality = getattr(ds, 'Modality', None)
    body_part = getattr(ds, 'BodyPartExamined', None)
    series_desc = getattr(ds, 'SeriesDescription', None)
    
    return {
        'modality': modality,
        'body_part': body_part,
        'series_description': series_desc
    }

Modality Mapping

Map DICOM codes to human-readable names:

Body Part Detection

Extract body part from text:

def detect_body_part(text):
    text_upper = text.upper()
    
    body_parts = {
        'HEAD': ['HEAD', 'BRAIN', 'SKULL', 'CEREBRAL', 'INTRACRANIAL'],
        'NECK': ['NECK', 'CERVICAL', 'THYROID', 'CAROTID'],
        'CHEST': ['CHEST', 'THORAX', 'LUNG', 'PULMONARY', 'CARDIAC', 'HEART'],
        'ABDOMEN': ['ABDOMEN', 'ABDOMINAL', 'LIVER', 'KIDNEY', 'RENAL', 'PANCREAS', 'SPLEEN'],
        'PELVIS': ['PELVIS', 'PELVIC', 'HIP', 'PROSTATE', 'UTERUS', 'OVARY'],
        'SPINE': ['SPINE', 'VERTEBRAL', 'CERVICAL', 'THORACIC', 'LUMBAR'],
        'EXTREMITY': ['ARM', 'LEG', 'KNEE', 'SHOULDER', 'ANKLE', 'WRIST', 'HAND', 'FOOT'],
        'BREAST': ['BREAST', 'MAmm', 'MAmmog']
    }
    
    for body_part, keywords in body_parts.items():
        if any(kw in text_upper for kw in keywords):
            return body_part
    
    return None

Contrast Detection

Determine if contrast is used:

def detect_contrast(text):
    text_upper = text.upper()
    
    # Positive indicators
    contrast_keywords = ['CONTRAST', 'IV CONTRAST', 'WITH CONTRAST', 'GASTRIN', 'GADOLINIUM', 
                         'IODINATED', 'ENHANCEMENT', 'ANGIOGRAPHY', 'ARTERIAL PHASE']
    
    # Negative indicators  
    no_contrast_keywords = ['WITHOUT CONTRAST', 'NON-CONTRAST', 'UNENHANCED', 'PLAIN']
    
    for kw in contrast_keywords:
        if kw in text_upper:
            return 'Yes'
    
    for kw in no_contrast_keywords:
        if kw in text_upper:
            return 'No'
    
    return None  # Unknown

Output Format

Return detection results in structured format:

{
  "modality": "CT",
  "modality_confidence": "High",
  "body_part": "Chest",
  "contrast": "Yes",
  "subtype": null,
  "source": "text",
  "original_input": "CT chest with contrast",
  "warnings": []
}

Confidence Levels

Handling Ambiguous Cases

When input is ambiguous:

1. Request clarification: Ask user for more specific information
2. List possibilities: Provide options if multiple modalities match
3. Use context: Consider clinical context if available
4. Defer to user: When in doubt, ask rather than guess

Example response for ambiguous input:

The input "chest imaging study" is ambiguous. Please clarify:

1. CT chest with contrast
2. Chest X-ray (PA/lateral)
3. Chest ultrasound
4. PET/CT chest

Which modality do you mean?

Common Abbreviations

Related Skills

• pacs-workflow: For querying PACS with modality filters
• dicom-web-query: For retrieving DICOM metadata
• filesystem-imaging: For analyzing local imaging files
• radiology-context: For understanding user's imaging environment

Examples

Example 1: Text Input

Input: "CT abdomen with contrast" Output:

{
  "modality": "CT",
  "body_part": "Abdomen",
  "contrast": "Yes"
}

Example 2: DICOM File

Input: DICOM file with Modality=MR, BodyPartExamined=BRAIN Output:

{
  "modality": "MRI",
  "body_part": "Brain",
  "source": "DICOM header"
}

Example 3: Ambiguous Input

Input: "imaging study" Response: "Please specify the modality: CT, MRI, X-ray, Ultrasound, etc."