A deep learning based ultrasound diagnostic tool driven by 3D visualization of thyroid nodules

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Static ultrasound images present limitations for accurate thyroid nodule diagnosis.
• Computer-assisted diagnostic tools require enhancement for clinical utility.

Purpose of the Study:

• To develop and validate a novel diagnostic tool, Thyroid Nodules Visualization (TNVis), for thyroid nodule diagnosis.
• To leverage dynamic ultrasound video and 3D deep learning for improved diagnostic accuracy.

Main Methods:

• A two-stage deep learning framework incorporating 3D visualization was developed.
• The framework, TNVis, was trained on 4569 cases from a multicenter study.
• Diagnostic validation was performed using data from seven hospitals.

Main Results:

• TNVis achieved a Dice similarity coefficient of 0.90 in internal testing.
• External validation showed TNVis significantly improved radiologists' performance (AUC 0.79) compared to unassisted (AUC 0.66) or partially assisted (AUC 0.72) diagnosis.
• The tool demonstrated a statistically significant improvement (p < 0.001).

Conclusions:

• The TNVis-assisted diagnostic strategy enhances radiologists' diagnostic capabilities.
• TNVis mimics clinical diagnostic procedures and provides objective 3D nodule representation.
• This facilitates precise, personalized diagnosis and treatment planning for thyroid nodules.