Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Thyroid nodule management requires improved diagnostic strategies.
• Current methods may lead to unnecessary procedures.
• Artificial intelligence (AI) offers potential for enhanced diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a deep-learning AI model (ThyNet) for thyroid nodule diagnosis.
• To assess ThyNet's ability to improve radiologists' diagnostic performance.
• To investigate the potential of ThyNet in reducing unnecessary fine needle aspirations.

Main Methods:

• ThyNet was trained on 18,049 images from 8,339 patients and tested on 4,305 images from 2,775 patients across multiple hospitals.
• Diagnostic performance was compared between ThyNet and 12 radiologists.
• A ThyNet-assisted strategy was developed and tested in simulated and real-world clinical settings.

Main Results:

• ThyNet achieved a higher area under the receiver operating characteristic curve (AUROC) (0.922) than radiologists (0.839).
• The ThyNet-assisted strategy improved radiologists' pooled AUROC from 0.837 to 0.875 for image review and from 0.862 to 0.873 in a clinical setting.
• Unnecessary fine needle aspirations decreased from 61.9% to 35.2%, with a slight decrease in missed malignancies from 18.9% to 17.0%.

Conclusions:

• The ThyNet-assisted strategy significantly enhances radiologists' diagnostic accuracy for thyroid nodules.
• This AI-driven approach can effectively reduce the rate of unnecessary fine needle aspirations.
• Integrating AI like ThyNet holds promise for optimizing thyroid nodule management.