Back to results
US 10,762,637 B2Granted
Vascular segmentation using fully convolutional and recurrent neural networks
AssigneeSiemens Healthcare GmbH
InventorsMehmet Akif Gulsun, Yefeng Zheng, Puneet Sharma, Vivek Kumar Singh, Tiziano Passerini
Life of this patent
Estimating the term…
Abstract
Systems and methods are provided for automatic segmentation of a vessel. A sequence of image slices containing a vessel is acquired. Features maps are generated for each of the image slices using a trained fully convolutional neural network. A trained bi-directional recurrent neural network generates a segmented image based on the feature maps.
Claims14
A method for segmenting a plurality of images, the method comprising:
- acquiring, by an image processor, the plurality of images, wherein the plurality of images comprise sequential image slices of a vessel along a centerline of the vessel, wherein each sequential image slice is centered at a respective sampling point and aligned orthogonally with a tangent direction to the centerline at the respective sampling point;
- generating, by the image processor, a feature map for each of the plurality of images using a machine-trained fully convolutional network;
- and
- generating, by the image processor, segmented images for each of the plurality of images based on the feature maps for the plurality of images using a bi-directional long short-term memory network comprising at least a forward direction long short-term memory layer and a backward direction long short-term memory layer that each input a respective feature map, wherein the outputs of the forward direction long short-term memory layer and the backward direction long short-term memory layer are averaged to generate a segmented image, wherein each of the segmented images includes labels for different types of tissues, obstructions, calcifications, and foreign objects.
1 more element
A method for segmenting a three-dimensional image, the method comprising:
- acquiring, by an image processor, the three-dimensional image;
- identifying a centerline of a vessel through the three-dimensional image;
- acquiring, by the image processor, a plurality of image slices at sampling points along the centerline, wherein each image slice is centered at a respective sampling point and aligned orthogonally with a tangent direction to the centerline at the respective sampling point;
- generating, by the image processor, with a machine-trained fully convolutional network, feature maps for each of the plurality of images slices;
- generating, by the image processor, segmented images for each of the plurality of images slices based on the feature maps with a bi-directional long short-term memory network comprising at least a forward direction long short-term memory layer and a backward direction long short-term memory layer that each input a respective feature map, wherein the outputs of the forward direction long short-term memory layer and the backward direction long short-term memory layer are averaged to generate the segmented images, each of the segmented images including labels for different types of tissues, obstructions, calcifications, and foreign objects;
- and
- generating, by the image processor, a model from the segmented images.
4 more elements
An imaging system for segmenting images for medical applications, the imaging system comprising:
- a medical image scanner configured to acquire a sequential series of two-dimensional cross-sectional images along a centerline of a vessel, wherein each two-dimensional cross-sectional image is centered at a respective sampling point and aligned orthogonally with a tangent direction to the centerline at the respective sampling point;
- and
- an image processor configured as a machine-trained fully convolutional network, the machine-trained fully convolutional network configured to generate a feature map for each of the plurality of two dimensional cross-sectional images, the image processor further configured as a bi-directional long short-term memory network comprising at least a forward direction long short-term memory layer and a backward direction long short-term memory layer that each input a respective feature map, wherein the outputs of the forward direction long short-term memory layer and the backward direction long short-term memory layer are averaged to generate segmented images for each of the two-dimensional cross-sectional images based on the feature maps, each of the segmented images including labels for different types of tissues, obstructions, calcifications, and foreign objects.
Dependencies are inferred from the claim text.