Category |
Assignment |
Subject |
Science |
University |
___ |
Module Title |
Healthcare Informatics |
Count Words: |
1500 words |
Academic year: |
2025 |
Learning Outcomes: Discuss An AI Application For Medical Imaging
- Understand clinic use
- Know dataset quality and diversity
- Explore performance metrics
- Know validation strategies
- Find clinic evaluation
- Apply ethical and regulatory requirements
- Deploy an application in the real world
Requirements
- Each student will select one AI application for medical imaging
- Evaluate and summarize the application based on the 7 key points in the previous slide
- If the information of a key point is not available yet, you should discuss how it can be done with references to relevant AI applications.
- Word limit: max 1500 words; figures, graphs, tables, illustrations, in-text citations, references, etc. are not counted for the limit; penalty applies if exceeding the limit
Requirements
- uAI recommended – demo access is provided
- Other AI applications acceptable
- Address the 7 key points given in the guidelines
- Rephrase essential concepts to show your understanding, e.g. performance metrics
- Use section titles as recommended: The scope of application, Training datasets, Performance metrics, validations, Clinical evaluation outcome, Ethical and regulatory, Deployment and improvement, Reference
1. The Scope Of Application
- Imaging modality, e.g. CT, MRI, X-ray, Angio etc.
- Patient population, pediatric vs adult, inpatient vs outpatient etc.
- Use conditions: e.g. lesion detection, scoliosis analysis, bone age assessment, fractures, pneumonia, PE, ICH, etc.

2. Dataset Quality And Diversity
- Number of image, age-range, race-range, to represent the target population
- Annotation quality, e.g. ground truth was established by experts (radiologists)
- Bias assessment, e.g. demographic variation, equipment differences
3. Performance Metrics

4. Validation Strategy
- Internal validation: the same dataset was used for the training and validation
- External validation: the training dataset is different from the validation dataset, e.g. from different institutions or populations
- Cross-validation: the dataset was split into multiple parts and training was done multiple times

5. Clinical Evaluation
- Expert studies: AI performance vs the experts (radiologists)
- Workflow integration: Evaluate how AI fits the clinical workflows
- Decision support: Assess and quantify what AI improves/brings to the clinical practices, e.g., reduce workload, improve workflow, provide better confidence, etc.

6. Regulatory And Ethical Considerations
- Compliance: HAS (Singapore), FDA (USA), NMPA (China) CE Mark (European), Number of image, age-range, race-range, to represent the target population
- Transparency: How are the AI decisions understood?
- Data privacy: compliance with PDDA, HIPPA, GDPR, PIPL, etc
7. Real-World Deployment
- Pilot studies: deploy the application in a limited clinical setting
- Monitoring: Track performance over time and across different settings
- Feedback loop: Incorporate clinician feedback for continuous improvements

8. References
Mendeley
- APA 7th Edition
- In-text citations whenever appropriate to avoid plagiarism
- A list of references at the end
- The words of references and in-text citations are excluded from the word limit
- Recommendation: Use Mendeley reference manager or Endnote to help you manage references; Mendeley reference manager is free.
Grading Rubric – Healthcare Informatic Assignment
