Roadmap for Representative Artificial Intelligence Models for Thyroid Cancer

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Artificial intelligence (AI) models are increasingly used in medical diagnostics, including thyroid cancer.
• Current AI models often exhibit poor generalizability due to limitations in training and validation datasets.
• This lack of generalizability hinders the clinical utility and fairness of AI tools in thyroid cancer management.

Purpose of the Study:

• To address the generalizability issue of AI models in thyroid cancer.
• To propose a structured approach for developing more robust and clinically useful AI models.
• To enhance the fairness and reliability of AI applications in oncology.

Main Methods:

• Development of a four-stage roadmap for AI model creation in thyroid cancer.
• Emphasis on representative data collection and curation.
• Inclusion of contextual variables and robust training/validation strategies.
• Incorporation of post-deployment monitoring for continuous improvement.

Main Results:

• The proposed roadmap provides a framework to mitigate bias in AI datasets.
• It facilitates the development of AI models that are more representative of diverse patient populations.
• Implementation of the roadmap is expected to improve the fairness and clinical applicability of AI in thyroid cancer.

Conclusions:

• Addressing data representativeness and model design is crucial for AI generalizability in thyroid cancer.
• The proposed roadmap offers a systematic strategy to enhance the fairness and clinical utility of AI tools.
• Future AI development in oncology should prioritize these principles for reliable and equitable patient care.