Establishing a Validation Infrastructure for Imaging-Based Artificial Intelligence Algorithms Before Clinical...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Validation

Background:

• FDA-cleared artificial intelligence (AI) algorithms require local evaluation before clinical integration.
• Ensuring AI accuracy and generalizability is vital for patient safety and health equity.
• Challenges like data privacy and intellectual property hinder external AI validation.

Purpose of the Study:

• To propose solutions for developing efficient, customizable, and cost-effective external validation infrastructures for AI models.
• To outline steps for establishing AI inferencing infrastructures outside clinical systems for local performance assessment.
• To promote an evidence-based approach for adopting AI models in healthcare.

Main Methods:

• Developing external validation infrastructures for AI models.
• Establishing AI inferencing infrastructures separate from clinical systems.
• Examining local performance of AI algorithms prior to implementation.

Main Results:

• Proposed strategies address challenges in AI model validation.
• A framework for local AI performance assessment is presented.
• The approach facilitates evidence-based AI adoption.

Conclusions:

• Robust local validation infrastructures are essential for safe and equitable AI integration in healthcare.
• External validation frameworks can overcome data privacy and IP concerns.
• Implementing these infrastructures enhances radiology workflows and patient outcomes.