AI and Radiology Second-Read: Where Algorithmic Triage Helps and Where It Hurts

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-healthcare-AI-and-radiology-second-read-r13a6-adults

1. What is the primary clinical function of FDA-cleared CADt tools like Aidoc or Viz.ai in radiology departments?

   1. To provide autonomous diagnosis of imaging studies without radiologist oversight
   2. To reorder the worklist so that suspected emergencies are read first
   3. To replace radiologists for routine, non-urgent studies
   4. To generate final radiology reports that require no radiologist review

2. Which of the following best describes the risk of automation bias in AI-assisted radiology?

   1. The radiologist trusts the AI flag too much and may miss errors
   2. The radiologist becomes more thorough due to AI confirmation
   3. AI reduces the overall number of studies read per day
   4. Automation bias improves inter-rater reliability

3. According to current FDA clearance parameters, which capability are radiology AI tools CLEARED to perform?

   1. Issuing final reports without radiologist sign-off
   2. Full replacement of the radiologist's clinical judgment
   3. Triage and prioritization of studies on a worklist
   4. Autonomous diagnostic interpretation of any imaging finding

4. A radiology department implements CADt software that flags suspected strokes. What is the appropriate workflow habit the radiologist should develop?

   1. Read every study at the same level of attention regardless of AI flags
   2. Read only the AI-flagged studies to save time
   3. Prioritize non-urgent studies over flagged emergencies
   4. Skip the AI-flagged studies since the algorithm already confirmed normality

5. An AI tool trained to detect intracranial hemorrhage (ICH) reviews a CT scan that contains an missed lung cancer. What limitation does this scenario illustrate?

   1. The AI was not powered by sufficient GPU resources
   2. The AI cannot read studies from scanners it wasn't trained on
   3. The AI can only detect findings within its narrow training scope
   4. The AI intentionally withheld the finding for legal reasons

6. A radiologist relies on AI-generated report skeletons without modification. What risk does this introduce?

   1. The radiologist may accept AI-suggested language uncritically
   2. The report skeleton eliminates the need for any dictation
   3. The report will be automatically cosigned by the AI
   4. The AI will generate a legally binding diagnosis

7. What is the current standard of care regarding AI tools in radiology interpretation?

   1. The hospital rather than the radiologist bears diagnostic responsibility
   2. The AI tool is legally considered the co-signer of the report
   3. The radiologist's independent read remains the standard of care
   4. AI recommendations supersede the radiologist's judgment in emergencies

8. Which statement accurately describes what FDA-cleared radiology AI tools can do?

   1. They can provide autonomous reads without any radiologist involvement
   2. They can compare today's study against prior studies in the same worklist
   3. They can interpret any imaging protocol regardless of training
   4. They can legally sign reports as the interpreting physician

9. A CADt system flags a study as suspected pulmonary embolism. What does this flag represent functionally?

   1. A recommendation for a second look that does not override the radiologist
   2. An automated report ready for immediate submission
   3. A directive to skip reading that specific study
   4. A confirmed diagnosis of pulmonary embolism

10. Why might an AI system fail to detect a finding in a study from a scanner protocol it wasn't trained on?

    1. Scanner protocols have no impact on AI performance
    2. The AI intentionally ignores protocols it doesn't like
    3. The scanner manufacturer blocked the AI from accessing the images
    4. The AI was only trained on specific protocol parameters and may not generalize

11. When an AI tool is integrated into a radiology worklist, what is its primary workflow optimization?

    1. Eliminating the need for prior study comparison
    2. Prioritizing cases based on suspected urgency
    3. Automatically distributing studies to other departments
    4. Reading studies autonomously to reduce radiologist workload

12. What should a radiologist do if the AI misses a finding that the radiologist also missed, but the AI wasn't trained to flag that finding type?

    1. Blame the AI manufacturer for the miss
    2. Consider the AI at fault since it failed to flag the finding
    3. Assume no liability since the AI didn't flag it
    4. Recognize that the standard of care remains the radiologist's read—the AI is a tool, not a co-signer

13. Which of the following is NOT a capability of current FDA-cleared radiology AI tools?

    1. Comparing current studies to prior studies in the worklist
    2. Replacing the radiologist's diagnostic interpretation
    3. Flagging suspected emergencies on the worklist
    4. Generating structured report skeletons from dictation

14. A radiology practice enables triage flagging in their AI worklist tool. What additional configuration is recommended to maintain diagnostic quality?

    1. Configure the habit to read every study with equal attention regardless of flags
    2. Only enable flagging for night shift studies
    3. Disable the flagging feature during weekday hours
    4. Set the AI to auto-dismiss flags after 30 seconds

15. A patient has a CT scan using a novel protocol not commonly used during the AI's training. What might happen when the AI processes this study?

    1. The AI will automatically adapt and perform optimally
    2. The protocol difference will improve the AI's accuracy
    3. The AI may produce unreliable results since it wasn't trained on this protocol
    4. The AI will refuse to process the study entirely