Applications of machine learning and deep learning to thyroid imaging: where do we stand?

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Endocrinology

Background:

• Ultrasonography (US) is crucial for thyroid nodule evaluation, but its interpretation is subjective, leading to variability.
• Operator-dependent US interpretation and interobserver variability contribute to unnecessary fine-needle aspirations (FNA) and surgeries.
• Computer-aided diagnosis (CAD) systems using AI aim to improve the accuracy and consistency of US feature interpretation for thyroid nodules.

Purpose of the Study:

• To provide a developmental overview of current AI-based CAD systems for thyroid nodules.
• To discuss future directions for AI in optimizing the personalized management of thyroid nodules.

Main Methods:

• Review of existing literature on AI-based CAD systems for thyroid nodule diagnosis.
• Analysis of the current capabilities and limitations of these AI systems.
• Exploration of emerging trends and future research avenues in AI for thyroid nodule management.

Main Results:

• AI-based CAD systems show promise in enhancing the accuracy and consistency of thyroid US interpretation.
• These systems have the potential to significantly decrease the rate of unnecessary FNA procedures.
• Current AI systems are evolving towards more sophisticated applications in personalized thyroid nodule management.

Conclusions:

• AI-based CAD systems represent a significant advancement in the diagnostic approach to thyroid nodules.
• Further development is expected to lead to more personalized and optimized treatment strategies.
• AI integration can improve clinical decision-making, reducing invasive procedures and healthcare costs.