Optimizing multiparametric ultrasound (mUS) combining B-mode, SWEI, ARFI,QUS Midband Fit for Prostate Gland Auto segmentation

Becky Arbiv      Rohin Maganti      Erik Tran

rivka.arbiv@duke.edu     rohin.maganti@duke.edu     erik.tran@duke.edu

Paper PDF

Prostate cancer is one of the most commonly diagnosed cancers and one of the leading causes of cancer death in men. Segmentation of the prostate is used to facilitate diagnostic and therapeutic treatment in a clinical setting. The task of segmenting the outer shell of the prostate gland is typically performed manually; however this process is time consuming and inconsistent. The goal of this work is two fold: (1)to develop a convolutional neural network (CNN) that performs semantic segmentation of the prostate based on B-mode images, and (2) to implement a physical layer into the network that will identify the imaging modality best suited for the task of segmentation. This will decrease the time needed to analyze data, while also increasing the reproducibility of segmented results, in addition to potentially decreasing the number of different modalities needed for prostate segmentation. We utilized a U-Net architecture with a trainable physical layer for this task resulting in a test accuracy of 0.9291.

Paper: Paper PDF

Code and Data: link to code here: B-Mode Segmentation and Physical Layer.