Deep Learning Ultrasound Computed Tomography with Sparse Transmissions

Ultrasound computed tomography (USCT) has been developed as a promising tool for breast cancer diagnosis and screening. One of the USCT reflection image reconstruction methods is synthetic aperture (SA), which can achieve isotropic high-resolution images. However, to reconstruct high-quality images, data acquisition strategies commonly involve full-matrix capture (FMC). FMC can lead to a time-consuming scanning process and heavy data-process demands, thus resulting in low clinical throughput. This paper proposes a framework for image restoration from sparse transmissions (TX) based on a deep convolutional neural network and demonstrates its performance with breast imaging in vivo. The expected image quality can be restored from the trained network with images reconstructed from only 8 TXs. To generate high-resolution reference images from the FMC data for training, a modified SA method was developed, which considered the sound speed difference between the coupling medium and the object. The corresponding network input images were reconstructed using sub-sampled data from 8 TXs distributed uniformly. The U-Net deep learning network was trained to remove the artifacts in reflection images reconstructed from sparse TX. The performance of the proposed framework was evaluated in terms of the structural similarity index measurement (SSIM) and the peak signal-to-noise ratio (PSNR). The results demonstrated that the trained network was capable of enhancing the image quality. The proposed framework provides an efficient approach for the reconstruction of high-quality reflection images from sparse data acquisition scheme, which can also benefit other ultrasound imaging applications by accelerating data acquisition and image reconstruction process.