Reconstruction of reflection ultrasound computed tomography with sparse transmissions using conditional generative...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Ultrasound computed tomography (UCT) shows promise for breast cancer diagnosis and screening.
• Synthetic aperture imaging in UCT provides high-resolution anatomical images.
• Full data acquisition for UCT is time-consuming, and sparse data can degrade image quality with traditional methods.

Purpose of the Study:

• To develop a deep learning framework, UCT-GAN, for efficient reconstruction of reflection UCT images from sparse transmission data.
• To address the challenge of reduced image quality in UCT when using sparse data acquisition strategies.

Main Methods:

• A conditional generative adversarial network (UCT-GAN) was designed for reflection UCT image reconstruction.
• The framework was evaluated using in vivo breast imaging data with a sparse transmission strategy (8 transmissions).
• Performance was quantitatively assessed using Peak Signal-to-Noise Ratio (PSNR), Normalized Mean Square Error (NMSE), and Structural Similarity Index Measurement (SSIM).

Main Results:

• UCT-GAN successfully generated high-quality reflection UCT images using only 8 transmissions, comparable to images from 512 transmissions.
• Quantitative metrics demonstrated UCT-GAN's superior performance over other methods like RED-GAN, DnCNN-GAN, and BM3D.
• Specifically, with 8-transmission sparse data, PSNR reached 29.52 dB and SSIM reached 0.7619.

Conclusions:

• The proposed UCT-GAN framework efficiently reconstructs high-quality reflection UCT images from sparse data.
• This deep learning approach has the potential to significantly reduce UCT scanning time for breast imaging.
• The method shows promise for integration into clinical UCT imaging systems for improved early breast cancer diagnosis.