Ultrasound segmentation using U-Net: learning from simulated data and testing on real data

Abstract

Segmentation of ultrasound images is an essential task in both diagnosis and image-guided interventions given the ease-of-use and low cost of this imaging modality. As manual segmentation is tedious and time consuming, a growing body of research has focused on the development of automatic segmentation algorithms. Deep learning algorithms have shown remarkable achievements in this regard; however, they need large training datasets. Unfortunately, preparing large labeled datasets in ultrasound images is prohibitively difficult. Therefore, in this study, we propose the use of simulated ultrasound (US) images for training the U-Net deep learning segmentation architecture and test on tissue-mimicking phantom data collected by an ultrasound machine. We demonstrate that the trained architecture on the simulated data is transferrable to real data, and therefore, simulated data can be considered as an alternative training dataset when real datasets are not available. The second contribution of this paper is that we train our U-Net network on envelope and B-mode images of the simulated dataset, and test the trained network on real envelope and Bmode images of phantom, respectively. We show that test results are superior for the envelope data compared to B-mode image.

Deep learning for emergency ascites diagnosis using ultrasonography images

“…The 2 × 2 convolution layer kernel was the same as that of the contracting path and upsampling layer. The U‐net included 23 total convolutional layers, the last of which exhibited a 1 × 1 convolution kernel, and was trained over 200 epochs with a batch size of 16 40,42 . A binary cross‐entropy loss function was applied with a learning rate of 0.0001, an Adam optimizer, and a He‐normal weight initializer function 40,48,49 .…”

“…The U-net included 23 total convolutional layers, the last of which exhibited a 1 × 1 convolution kernel,and was trained over 200 epochs with a batch size of 16. 40,42 A binary cross-entropy loss function was applied with a learning rate of 0.0001, an Adam optimizer, and a Henormal weight initializer function. 40,48,49 All experiments were performed on an Intel Core i5 (12G) computer with a single GeForce GTX 1080 Ti GPU.…”

“…This is particularly beneficial for first‐response and emergency situations, where access to specialists is limited. The U‐net is a DL model with an encoder–decoder architecture that was first proposed in 2015 40–42 . U‐nets have previously been applied to biomedical image processing problems such as liver and tumor segmentation in computed tomography (CT) scans, mass and calcification detection in digital mammograms, and the segmentation of skin lesions 40,42–45 .…”

“…The Unet is a DL model with an encoder-decoder architecture that was first proposed in 2015. [40][41][42] U-nets have previously been applied to biomedical image processing problems such as liver and tumor segmentation in computed tomography (CT) scans, mass and calcification detection in digital mammograms, and the segmentation of skin lesions. 40,[42][43][44][45] These models are composed of multiple layers and thus learn different hierarchical features in each iteration.…”

“…[40][41][42] U-nets have previously been applied to biomedical image processing problems such as liver and tumor segmentation in computed tomography (CT) scans, mass and calcification detection in digital mammograms, and the segmentation of skin lesions. 40,[42][43][44][45] These models are composed of multiple layers and thus learn different hierarchical features in each iteration. 46 Scholars have recently improved on the basic U-net structure, proposing more powerful frameworks such as UNet++, UNET 3+, and H-DenseUNet.…”