Deep neural network for multiparametric ultrasound imaging of prostate cancer

This study presents a deep neural network (DNN) for generating a multiparametric ultrasound (mpUS) volume of the prostate by combining data from acoustic radiation force impulse (ARFI) imaging, shear wave elasticity imaging (SWEI), B-mode imaging, and quantitative ultrasound-midband fit (QUS-MF). The DNN was trained using in vivo data to maximize the contrast-to-noise ratio between prostate cancer and healthy tissue. The network was evaluated in a prostate phantom, where the DNN was shown to increase the CNR of lesions as well as the CNR between the peripheral zone and the background. In a test in vivo dataset, the DNN improved the visibility of a histology-confirmed lesion. These findings suggest that deep learning may be a promising approach for providing enhanced imaging guidance during a biopsy of the prostate.

Automated summary

This study introduces a deep neural network (DNN) to generate a multiparametric ultrasound (mpUS) volume of the prostate by combining different ultrasound imaging techniques. The DNN was trained using in vivo data to improve contrast-to-noise ratio and was shown to enhance lesion visibility in both phantom and in vivo datasets. These results suggest that deep learning could offer improved imaging guidance for prostate biopsies.