Deep learning approach for the segmentation of aneurysmal ascending aorta

Abstract

Diagnosis of ascending thoracic aortic aneurysm (ATAA) is based on the measurement of the maximum aortic diameter, but size is not a good predictor of the risk of adverse events. There is growing interest in the development of novel imagederived risk strategies to improve patient risk management towards a highly individualized level. In this study, the feasibility and efficacy of deep learning for the automatic segmentation of ATAAs was investigated using UNet, ENet, and ERFNet techniques. Specifically, CT angiography done on 72 patients with ATAAs and different valve morphology (i.e., tricuspid aortic valve, TAV, and bicuspid aortic valve, BAV) were semi-automatically segmented with Mimics software (Materialize NV, Leuven, Belgium), and then used for training of the tested deep learning models. The segmentation performance in terms of accuracy and time inference were compared using several parameters. All deep learning models reported a dice score higher than 88%, suggesting a good agreement between predicted and manual ATAA segmentation. We found that the ENet and UNet are more accurate than ERFNet, with the ENet much faster than UNet. This study demonstrated that deep learning models can rapidly segment and quantify the 3D geometry of ATAAs with high accuracy, thereby facilitating the expansion into clinical workflow of personalized approach to the management of patients with ATAAs.

Deep Learning-Based Methods for Prostate Segmentation in Magnetic Resonance Imaging

“…The first model considered in this study was UNet, which has been adopted in several image delineation processes [45]. ENet [43], and ERFNet [44] have been implemented for the segmentation process in self-driving cars, and successfully used in lung and aorta segmentation tasks [48,49]. Specifically, they were used for the segmentation of HRCT images characterized by a very high number of slices for each study (about 600 and 450 slices for the lung and aorta studies, respectively).…”

“…Specifically, they were used for the segmentation of HRCT images characterized by a very high number of slices for each study (about 600 and 450 slices for the lung and aorta studies, respectively). Authors used 32 patients' studies for the parenchyma extraction process [49], and 72 studies for the aorta segmentation process [48]. In this study, only 85 studies were used considering that each patients' image dataset consists of about 40 slices.…”

“…Three different deep learning models including UNet [45], ENet [43], and ERFNet [44] were investigated to account for accurate prostate segmentation, fast training time, low hardware requirements for inference, and low training data requirements. Specifically, UNet was modified to improve segmentation accuracy, as reported in [48,49]. Briefly, (i) 3 × 3 convolutions were replaced by 5 × 5 convolutions, (ii) zero padding was used to ensure that the size of the output feature maps was the same as the input size, and (iii) an input size of 512 × 512 with 32 filters was used on the first contraction path layer, with doubling of feature maps after each max pool and stopping at 256 feature maps and 2D size of 64 × 64.…”

“…Briefly, (i) 3 × 3 convolutions were replaced by 5 × 5 convolutions, (ii) zero padding was used to ensure that the size of the output feature maps was the same as the input size, and (iii) an input size of 512 × 512 with 32 filters was used on the first contraction path layer, with doubling of feature maps after each max pool and stopping at 256 feature maps and 2D size of 64 × 64. Table 1 in [43] and Table 2 in [44] for the description of their architecture), they were mainly applied in self-driving cars to compensate for limited hardware availability maintaining high accuracy and successfully used in two biomedical segmentation issues [48,49], that is, in the segmentation of high resolution computed tomography (HRCT) images characterized by a slice thickness much lower than that of the T2 weighted images of the prostate studies. This means that the number of slices of each patients' study was much greater than in this study.…”

In-Series U-Net Network to 3D Tumor Image Reconstruction for Liver Hepatocellular Carcinoma Recognition

“…In radiomics studies [3], all observations underline the need for automatic and reliable tools dedicated to tumor segmentation in order to finely characterize liver cancer. However, automatic segmentation [4][5][6][7][8] of liver tumor is challenging not only due to the highly variable shape of liver tumors but also because of the similar intensity values of nearby liver parenchyma.…”