[Reconstruction of elasticity modulus distribution base on semi-supervised neural network]

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Deep Learning

Context:

• Ultrasound elastography aims to reconstruct tissue elasticity modulus distribution.
• Current deep learning methods often rely on simulated data, lacking real-world complexity.
• In-vivo ultrasound data presents unique challenges not fully captured by simulations.

Purpose:

• To improve the accuracy of deep learning-based ultrasound elastography reconstruction.
• To introduce a semi-supervised approach utilizing real in-vivo displacement data for training.
• To compare the performance of the proposed semi-supervised model against a fully supervised model.

Summary:

• A semi-supervised deep learning model was trained using real in-vivo ultrasound displacement data.
• Phantom experiments showed the semi-supervised model achieved ~3% error, outperforming the fully supervised model's ~5% error.
• The semi-supervised model demonstrated reduced prediction error areas when processing real in-vivo data.

Impact:

• The study validates the effectiveness of incorporating real in-vivo data into semi-supervised learning for elastography.
• Provides a more practical and accurate deep learning approach for reconstructing elastic distributions from ultrasound data.
• Offers new insights for advancing deep learning applications in medical imaging and biomechanics.