Fully automatic deep learning trained on limited data for carotid artery segmentation from large image volumes

Abstract

Background: The objectives of this study were to develop a 3D convolutional deep learning framework (CarotidNet) for fully automatic segmentation of carotid bifurcations in computed tomography angiography (CTA) images and to facilitate the quantification of carotid stenosis and risk assessment of stroke.Methods: Our pipeline was a two-stage cascade network that included a localization phase and a segmentation phase. The network framework was based on the 3D version of U-Net, but was refined in three ways: (I) by adding residual connections and a deep supervision strategy to cope with the vanishing problem in back-propagation; (II) by adopting dilated convolution in order to strengthen the capacity to capture contextual information; and (III) by establishing a hybrid objective function to address the extreme imbalance between foreground and background voxels.Results: We trained our networks on 15 cases and evaluated their performance based on 41 cases from the MICCAI Challenge 2009 dataset. A Dice similarity coefficient of 82.3% was achieved for the test cases. Conclusions:We developed a carotid segmentation method based on U-Net that can segment tiny carotid bifurcation lumens from very large backgrounds with no manual intervention. This was the first attempt to use deep learning to achieve carotid bifurcation segmentation in 3D CTA images. Our results indicate that deep learning is a promising method for automatically extracting carotid bifurcation lumens.

State-of-the-art CT and MR imaging and assessment of atherosclerotic carotid artery disease: standardization of scanning protocols and measurements—a consensus document by the European Society of Cardiovascular Radiology (ESCR)

“…The contributions of ML in the context of carotid artery disease can be assigned to four broader categories. First, carotid artery segmentation [144][145][146][147], which is the basis for many secondary analyses, provides potential for more comprehensive analyses of vessel anatomy and pathology. For example, Tsakanikas et al proposed a U-net model to produce a 3D meshed model of the carotid bifurcation and branches using multispectral MR image series and reported an accuracy of 99.1% for lumen area [144].…”

Ability of weakly supervised learning to detect acute ischemic stroke and hemorrhagic infarction lesions with diffusion-weighted imaging

“…Conventionally, in a CNN that is designed for classification, the last convolution layer is usually followed by a fully connected layer to output the classification result (22). In this study, since the objective was to detect the lesion while the labels were the manual annotation denoting whether or not a slice contained a stroke lesion, we replaced the 4096-dimension fully connected layers in the original VGG-16 and ResNet-50 by a global average pooling layer followed by an output layer (i.e., a fully connected layer).…”

Group theoretic particle swarm optimization for multi-level threshold lung cancer image segmentation

“…The characteristic appearances of lung cancer images are the slow growth of a localized ground glass opacity and the rapid increment of a solid mass (8). And many promising technologies have been developed for detecting those appearance changes of medical image in segmentation tasks with remarkable performance, they range from traditional approaches, like thresholding, edgebased extraction, histogram-based bundling and watershed, region-based growing, to more superior methods, like active contour models, sparse representations, conditional and Markov random fields, graph cuts, deep learning-based algorithms (9)(10)(11)(12). Although there exist many excellent methods like deep learning approaches, they still have some inevitable issues, for example, the results heavily depend on complex model structure and parameter configuration.…”