引导式测验 · 硬 等级

医疗保健中的人工智能测验

通过预期用途、代表性患者和设备、人类监督、隐私以及为何监管状态针对特定用途来评估医疗保健人工智能。

相关引导路径人工智能医疗
问题 1 的 8

A model ranks emergency records for review. Why does success at ranking not establish its diagnostic accuracy?

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. A model ranks emergency records for review. Why does success at ranking not establish its diagnostic accuracy?
    • Ranking for triage and suggesting diagnoses are different intended uses requiring different evidence
    • Ranking urgent records is always a diagnosis
    • A model’s brand determines its clinical role
    • A benchmark covers every possible clinical workflow
  2. What is needed before assuming a model will work on a new patient group?
    • A higher confidence threshold only
    • Evaluation with representative data and relevant subgroups
    • A copy of the training loss
    • A product brochure
  3. An imaging aid is moved to a hospital using different scanners. Why should it be evaluated again?
    • Devices cannot affect medical images
    • A model automatically normalizes every source
    • Scanner differences can alter the inputs and performance in the new workflow
    • Workflow is unrelated to clinical use
  4. What can a retrospective result establish about future clinical use?
    • It guarantees safety in deployment
    • It proves the model generalizes to all hospitals
    • It replaces monitoring
    • It can support investigation but does not alone establish prospective readiness
  5. What does FDA device status apply to?
    • A particular device and intended use, not every related model or workflow
    • Every system from the same company
    • All models trained on medical data
    • Any assistant that mentions a diagnosis
  6. What should happen when a healthcare model is uncertain or unavailable?
    • Hide the uncertainty from clinicians
    • Provide a route for qualified human review
    • Make an automatic care decision anyway
    • Treat missing output as a negative result
  7. Which information needs protection in an AI health workflow?
    • Only patient names
    • Only images before processing
    • Inputs, logs, derived features, and outputs
    • Nothing if the model is hosted remotely
  8. What should be recorded when evaluating a healthcare AI system?
    • Only the hospital name
    • 只有最终的准确率数字
    • Only the patient-facing interface
    • Model, data, version, and context of use