Dual-View Thyroid Ultrasound Classification via Dual Knowledge Distillation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Thyroid ultrasound diagnosis commonly uses transverse and longitudinal views.
• Existing deep learning models often process these views independently, limiting diagnostic robustness.
• There's a need for methods that effectively model cross-view semantic consistency.

Purpose of the Study:

• To propose an uncertainty-weighted mixture-of-experts (UMoE) framework for dual-view thyroid ultrasound classification.
• To improve cross-view semantic consistency and diagnostic robustness in automated thyroid nodule analysis.
• To leverage dual knowledge distillation for enhanced classification performance.

Main Methods:

• A shared Vision Transformer backbone with dual-view and single-view branches.
• Cross-view knowledge distillation using uncertainty-refined patch tokens to align representations.
• Cross-level knowledge distillation transferring holistic semantic knowledge from dual-view to single-view branches.

Main Results:

• The UMoE framework consistently outperformed single-view and dual-view baselines in accuracy, F1-score, and AUC.
• Experiments on in-house and external datasets validated the model's superior performance.
• Ablation studies confirmed the effectiveness of the dual-view strategy and individual components.

Conclusions:

• The UMoE framework enhances cross-view consistency and predictive ability of single-view branches for accurate thyroid ultrasound classification.
• This clinically relevant framework offers a practical strategy for improving computer-aided diagnosis of thyroid nodules.
• The study demonstrates a significant advancement in dual-view learning for medical image analysis.