TSE-GAN: strain elastography using generative adversarial network for thyroid disease diagnosis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thyroid cancer incidence has tripled globally in 35 years.
• Strain elastography is a sensitive imaging technique for thyroid nodule evaluation.
• Current strain elastography methods face limitations in standardization and analysis.

Purpose of the Study:

• To develop an automated method for generating realistic thyroid strain elastograms.
• To address the limitations of existing strain elastography techniques.
• To improve the diagnostic accuracy of thyroid nodule assessment.

Main Methods:

• Proposed a novel conditional generative adversarial network (TSE-GAN).
• Utilized a global-to-local architecture for multi-scale feature extraction.
• Incorporated an adaptive deformable U-net structure for effective deformation.
• Introduced a Lab-based loss function for realistic elastogram generation.

Main Results:

• TSE-GAN successfully generated realistic thyroid strain elastograms.
• The proposed method demonstrated superior performance compared to state-of-the-art image translation techniques.
• Qualitative and quantitative assessments confirmed the clinical utility of generated elastograms.

Conclusions:

• TSE-GAN offers a valuable tool for clinical thyroid nodule diagnosis.
• The AI-driven approach enhances the diagnostic application of strain elastography.
• This method holds practical value in improving thyroid cancer detection.