RadQA: Radiology Report Question Answering Dataset

# RadQA: Radiology Report Question Answering
# Based on Soni et al., LREC 2022
# Paper: https://aclanthology.org/2022.lrec-1.672/
# Dataset: https://physionet.org/content/radqa/1.0.0/
#
# Task: Extractive question answering on clinical radiology reports
# Annotators read radiology reports and highlight answer spans that address
# clinical questions, classify answerability, and identify answer types.
#
# Annotation workflow:
# 1. Read the clinical question displayed above the report
# 2. Read the full radiology report carefully
# 3. Highlight the span(s) in the report that answer the question
# 4. Classify whether the question is answerable from the report
# 5. Identify the type of answer (finding, measurement, location, etc.)

annotation_task_name: "RadQA: Radiology Report Question Answering"
task_dir: "."

data_files:
  - sample-data.json
item_properties:
  id_key: "id"
  text_key: "report"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_task_name: "RadQA Answer Extraction"

annotation_instructions: |
  You are performing extractive question answering on clinical radiology reports.

  **Instructions:**
  1. Read the **clinical question** displayed above the report text.
  2. Carefully read the entire **radiology report**.
  3. **Highlight** the text span(s) in the report that answer the question.
     - Use "answer" for spans that directly and fully answer the question.
     - Use "partial-answer" for spans that provide relevant but incomplete information.
  4. Classify whether the question is **answerable** from the report.
  5. Identify the **answer type** (what kind of clinical information the answer conveys).
  6. Copy the extracted answer text into the answer text field.

  **Important Notes:**
  - Select the minimal span that answers the question — do not include unnecessary context.
  - If the answer appears in multiple places, highlight the most informative occurrence.
  - Clinical abbreviations are acceptable in answer spans.
  - If the report does not contain information to answer the question, mark it as "unanswerable."

annotation_schemes:
  - annotation_type: span
    name: answer_span
    description: "Highlight the text span(s) in the report that answer the clinical question"
    labels:
      - "answer"
      - "partial-answer"
    label_colors:
      "answer": "#4caf50"
      "partial-answer": "#ff9800"
    tooltips:
      "answer": "Text that directly and fully answers the question"
      "partial-answer": "Text that provides relevant but incomplete information toward answering the question"
    allow_overlapping: false

  - annotation_type: radio
    name: answerability
    description: "Can the clinical question be answered from this radiology report?"
    labels:
      - name: "answerable"
        tooltip: "The report contains sufficient information to fully answer the question"
      - name: "unanswerable"
        tooltip: "The report does not contain information needed to answer the question"
      - name: "partially-answerable"
        tooltip: "The report contains some relevant information but not enough for a complete answer"
    keyboard_shortcuts:
      "answerable": "a"
      "unanswerable": "u"
      "partially-answerable": "p"

  - annotation_type: radio
    name: answer_type
    description: "What type of clinical information does the answer convey?"
    labels:
      - name: "finding"
        tooltip: "A clinical observation or abnormality (e.g., opacity, mass, effusion)"
      - name: "measurement"
        tooltip: "A quantitative value (e.g., size, volume, Hounsfield units)"
      - name: "location"
        tooltip: "An anatomical location or spatial relationship"
      - name: "comparison"
        tooltip: "A comparison with prior studies or temporal change"
      - name: "recommendation"
        tooltip: "A suggested follow-up action or further workup"
      - name: "diagnosis"
        tooltip: "A diagnostic impression or differential diagnosis"
    keyboard_shortcuts:
      "finding": "f"
      "measurement": "m"
      "location": "l"
      "comparison": "c"
      "recommendation": "r"
      "diagnosis": "d"

  - annotation_type: text
    name: answer_text
    description: "Copy the extracted answer text here (the text you highlighted as answering the question)"

html_layout: |
  <div style="margin-bottom: 12px; padding: 12px; background: #e3f2fd; border-radius: 6px; border-left: 4px solid #1565c0;">
    <strong>Report Type:</strong> {{report_type}}<br/>
    <strong style="font-size: 1.1em;">Question:</strong>
    <span style="font-size: 1.1em; color: #1565c0;">{{question}}</span>
  </div>
  <div style="padding: 12px; border: 1px solid #ddd; border-radius: 6px; line-height: 1.8; font-family: 'Courier New', monospace; background: #fafafa;">
    {{report}}
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false