Moderator: Laurent Létourneau-Guillon

Foundation Models and Large Language Models in Radiology
An Tang

Radiologists analyze medical images and describe their content in text reports. Recent technical developments in deep learning models that connect images and text may facilitate radiology workflow. Foundation models refer to large-scale models trained on extensive multimodal datasets that incorporate various data types such as images, videos, audio, and text. They are designed to support many applications such as image recognition and language understanding. Large language models (LLM) are a narrower type of AI model trained on large amounts of text data to understand and generate human language. A well-known example is GPT. AI models are adaptable and can be “fine-tuned” to accomplish specialized tasks such as adopting a medical lexicon. Potential clinical applications of these models include automated medical image captioning, preliminary radiology report generation, impression creation, and generation of educational images.

At the end of this session, participants will be able to:  

• Explain the difference between foundation and large language models.
• Describe key concepts of foundation model and large language model architectures.
• Recognize clinical applications of these models in radiology.

Target Audience:

• Radiologist
• Resident
• Medical Student

CanMEDS:

• Medical Expert
• Scholar
• Professional

- - -

Ethical Considerations and Fairness in AI
Christopher Filippi

This talk with explore ethical considerations and fairness in the development of Artificial Intelligence in imaging and healthcare. Issues of representation and generalizability will be discussed. Mathematical fairness and tools to uncover bias in algorithms will be given. At the end of the talk, a discussion of quality metrics (Quality Assurance and Quality Improvement) will be addressed and potential solutions for healthcare networks.

At the end of this presentation, participants will be able to:

• Define bias, ethics, and fairness in the setting of artificial intelligence in the context of clinical healthcare decision making in diagnostic radiology.
• Demonstrate familiarity with concepts of mathematical fairness and tools that can be used to mitigate bias in artificial intelligence algorithmic development.
• Acquire an understanding of the generalizability of artificial intelligence in clinical settings and explore strategies for governance to ensure the quality, fairness, and ethical deployment of AI tools in healthcare.

Target Audience:

• Radiologist
• Resident
• Medical Student

CanMEDS:

• Medical Expert
• Communicator
• Collaborator
• Leader
• Health Advocate
• Professional

- - -

How to Implement an AI Tool in My Department
Jaron Chong

- - -

AI Roundtable