Moderator: Jaron Chong

Clinical Evaluation of AI Algorithm: bridging the gap between efficacy and effectiveness
Guillaume Herpe

This presentation addresses the critical need to bridge the gap between the efficacy of Artificial Intelligence (AI) algorithms in controlled settings and their effectiveness in real-world clinical environments. It starts by examining the current advancements in AI in radiology and their implications. The focus then shifts to the concept of efficacy, where AI algorithms are initially developed and tested under ideal conditions, which may not accurately represent their performance in diverse clinical settings.

The main discussion centers on the challenges of deploying AI in clinical environments, such as data variability, algorithmic bias, ethical concerns, and integration into existing workflows. We emphasize the importance of robust clinical validation, including multicentric trials and real-world data analysis, to ensure consistent performance across various patient populations and settings.

The presentation also highlights the need for effective collaboration between clinicians and AI systems, stressing the importance of clinician involvement in AI development and the necessity for AI transparency. Finally, a framework for the continuous evaluation and adaptation of AI algorithms in radiology is proposed, advocating for systems that learn and evolve with new data and changing healthcare dynamics.

By the end of this session, participants should be better able to:
- Differentiate the critical distinction between AI algorithm efficacy in controlled settings and effectiveness in diverse clinical environments
- Explore methodologies for robust clinical validation of AI algorithms, including multicentric trials and real-world data analysis techniques.
- Identify strategies for effective clinician-AI collaboration, emphasizing the importance of clinician involvement and transparent AI systems.

Click below to complete the evaluation for this session
SESSION EVALUATION

- - - 

AI Update 2024: An Introduction to Large Language Models and Vision-Language Models
Jaron Chong

This presentation will explore the latest advancements in large language models (LLMs) and vision-language models (VLMs), introducing both the fundamental technical developments that have brought them to the scientific foreground. We will demonstrate the expected range of functions that these models will offer, their unique barriers to application specific fine tuning and development, and preview the latest examples of applications utilizing these technologies, with respect to how they are applied to other industries and how developments there could soon translate to diagnostic radiology.

By the end of this session, participants should be better able to:
- Review Transformers architecture, self-supervised learning definitions, and LLM/VLM data training requirements
- Understand the present limitations of AI generated text and images
- Compare and contrast Large Language Models from Vision Language Models

Click below to complete the evaluation for this session
SESSION EVALUATION

- - -

Predicting the Long-Term Strategic Application and Impact of Large-Language Models and Vision-Language Models
An Tang, Alexandre Cadrin-Chênevert, Mark Cicero & Jaron Chong

This panel discussion will dive into the transformative potential of advanced AI models within radiology AI. Panelists will explore how large-language models and vision-language models could alter diagnostic processes, approaches to AI education in radiology, experiences in deploying real-world AI and resolving barriers to adoption. The panel will also offer audience members the opportunity to pose questions to the panel.

By the end of this session, participants should be better able to:
- Analyze the ways large-language models and vision-language models could alter administrative and diagnostic processes in radiology.
- Evaluate real-world experiences of deploying AI in radiology departments.
- Develop strategies for AI education in radiology, concerning large-language models and vision-language models

Click below to complete the evaluation for this session
SESSION EVALUATION