Deep Learning for Automated Liver Segmentation to Aid in the Study of Infectious Diseases in Nonhuman Primates

Deep-Learning-Based Whole-Lung and Lung-Lesion Quantification Despite Inconsistent Ground Truth: Application to Computerized Tomography in SARS-CoV-2 Nonhuman Primate Models

“…Scans were stratified such that scans from the same animal did not appear in the training/validation folds and the test fold. Since the FPN was previously proven superior to U-Net and V-Net for liver segmentation in CT images of NHPs [31] , we applied only the FPN for whole-lung segmentation. The example images shown in Figure 1 illustrate the whole lung segmented by the FPN compared to the manual masks in multiple animals.…”

“…A standard solution to this scale-variant issue is to use feature pyramids that enable a model to detect objects across a wide range of scales by scanning the model over both positions and pyramid levels. An FPN was recently used for liver segmentation [31] in whole-body CT images of NHPs, was effective for liver segmentation before and after exposure with different viruses (e.g., Ebola virus, Marburg virus, and Lassa virus), and achieved an average Dice coefficient of 94.77%. CNN training parameters and loss-function details were previously characterized [31] .…”

“…An FPN was recently used for liver segmentation [31] in whole-body CT images of NHPs, was effective for liver segmentation before and after exposure with different viruses (e.g., Ebola virus, Marburg virus, and Lassa virus), and achieved an average Dice coefficient of 94.77%. CNN training parameters and loss-function details were previously characterized [31] . This study extended the liver segmentation method to apply to whole-lung and lung-lesion segmentation.…”

Toward the determination of sensitive and reliable whole-lung computed tomography features for robust standard radiomics and delta-radiomics analysis in a nonhuman primate model of coronavirus disease 2019

“…For training purposes, a total of 64 whole-lung CT scans (reconstructed using a B kernel of crab-eating macaques with the same imaging protocols) were used. The automated organ segmentation method, based on the convolutional neural network (CNN), used in this work has been described before 60 . The feature pyramids network (FPN), which produces a multiscale feature representation in which all levels, even the high-resolution levels, are semantically strong, was used in this work.…”

“…The automated organ segmentation method, based on the convolutional neural network (CNN), used in this work has been described before. 60 The feature pyramids network (FPN), which produces a multiscale feature representation in which all levels, even the high-resolution levels, are semantically strong, was used in this work. The network was trained using input patches of size 64 × 64 × 64 voxels, which were randomly extracted from both lung and nonlung areas with equal numbers.…”