Systems and methods for brain identifier localization
Embodiments of the present disclosure provide systems and methods for brain identifier localization. The methods include obtaining an image of a brain; determining region identifier probability map(s) of the brain, landmark identifier probability map(s) of the brain, and plane identifier probability map(s) of the brain based on the image and a neural network model; determining a segmentation result of a cerebral cortex of the brain, landmark identifier(s) of the brain, and plane identifier(s) of the brain, respectively, based on the region identifier probability map(s), the landmark identifier probability map(s), and the plane identifier probability map(s); constructing a target coordinate system based on the landmark identifier(s) and the plane identifier(s); and determining landmark(s) of the cerebral cortex based on the segmentation result of the cerebral cortex, the target coordinate system, and/or the landmark identifier(s).
1 . A system for brain identifier localization, comprising a processor configured to direct the system to perform operations including:
obtaining an image of a brain;
determining at least one region identifier probability map of the brain, at least one landmark identifier probability map of the brain, and at least one plane identifier probability map of the brain based on the image and a neural network model, wherein a value of each pixel point in the at least one region identifier probability map represents a probability value that the pixel point belongs to at least one region identifier of the brain, a value of each pixel point in the at least one landmark identifier probability map represents a probability value that the pixel point belongs to at least one landmark identifier of the brain, and a value of each pixel point in the at least one plane identifier probability map represents a probability value that the pixel point belongs to at least one plane identifier of the brain;
determining a segmentation result of a cerebral cortex of the brain, the at least one landmark identifier of the brain, and the at least one plane identifier of the brain, respectively, based on the at least one region identifier probability map, the at least one landmark identifier probability map, and the at least one plane identifier probability map;
constructing a target coordinate system based on the at least one landmark identifier and the at least one plane identifier; and
determining at least one landmark of the cerebral cortex based on at least one of the segmentation result of the cerebral cortex, the target coordinate system, and the at least one landmark identifier.
2 . The system of claim 1 , wherein the at least one landmark identifier probability map includes an anterior commissure (AC) probability map and a posterior commissure (PC) probability map, and the at least one landmark identifier includes an AC landmark and a PC landmark.
3 . The system of claim 1 , wherein the determining a segmentation result of a cerebral cortex of the brain, the at least one landmark identifier of the brain, and the at least one plane identifier of the brain, respectively, based on the at least one region identifier probability map, the at least one landmark identifier probability map, and the at least one plane identifier probability map includes:
for each of the at least one landmark identifier probability map of the brain, determining a position of a pixel point corresponding to a maximum probability value in the landmark identifier probability map as a position of a landmark identifier of the at least one landmark identifier.
4 . The system of claim 1 , wherein the at least one plane identifier probability map includes a midsagittal plane (MSP) probability map, and the at least one plane identifier includes an MSP.
5 . The system of claim 1 , wherein the determining a segmentation result of a cerebral cortex of the brain, the at least one landmark identifier of the brain, and the at least one plane identifier of the brain, respectively, based on the at least one region identifier probability map, the at least one landmark identifier probability map, and the at least one plane identifier probability map includes:
for each of the at least one plane identifier of the brain, determining a target point set based on the plane identifier probability map; and
obtaining the plane identifier by fitting the target point set.
6 . The system of claim 5 , wherein for each of the at least one plane identifier of the brain, the determining a target point set based on the plane identifier probability map includes:
determining a set of pixel points in the plane probability map as the target point set,
a probability of each of the set of pixel points being larger than a preset threshold.
7 . The system of claim 5 , wherein for each of the at least one plane identifier of the brain, the obtaining the plane identifier by fitting the target point set includes:
obtaining the plane identifier by fitting, based on a random sampling consensus (RANSAC) algorithm, the target point set.
8 . The system of claim 7 , wherein the RANSAC algorithm includes:
performing multiple loops of the following operations:
a) determining a subset by randomly sampling the target point set;
b) obtaining a plane by fitting points in the subset; and
c) determining a sum of squares of distances between the plane and remaining points of the target point set except the subset; and
determining a plane corresponding to one loop with a minimum sum of squares of distances among the multiple loops as the plane identifier.
9 . The system of claim 1 , wherein the at least one landmark of the cerebral cortex includes at least one of a frontmost point of the cerebral cortex, a rearmost point of the cerebral cortex, a leftmost point of the cerebral cortex, a rightmost point of the cerebral cortex, an inferior point of the cerebral cortex, and superior point of the cerebral cortex.
10 . The system of claim 1 , wherein the determining at least one landmark of the cerebral cortex based on at least one of the segmentation result of the cerebral cortex, the target coordinate system, and the at least one landmark identifier includes:
determining the at least one landmark of the cerebral cortex based on a maximum point or a minimum point of the cerebral cortex along one direction of three coordinate axes of the target coordinate system; or
determining a maximum point or a minimum point of the cerebral cortex that passes through the at least one landmark identifier and is parallel to one direction of three coordinate axes of the target coordinate system as the at least one landmark of the cerebral cortex.
11 . The system of claim 1 , wherein the neural network model is a multi-task model including a shared network layer and three branch network layers, and the three branch network layers include a first branch network layer, a second branch network layer, and a third branch network layer.
12 . The system of claim 11 , wherein the first branch network layer is configured to segment the brain, and output the at least one region identifier probability map.
13 . The system of claim 11 , wherein the second branch network layer is configured to locate the at least one landmark identifier of the brain, and output the at least one landmark identifier probability map.
14 . The system of claim 11 , wherein the third branch network layer is configured to locate the at least one plane identifier of the brain, and output the at least one plane identifier probability map.
15 . The system of claim 11 , wherein the neural network model is generated through a training process including:
obtaining training sample images and gold standard images corresponding to each of the training sample images, the gold standard images corresponding to each of the training sample images including at least one region identifier gold standard image, at least one landmark identifier gold standard image, and at least one plane identifier gold standard image corresponding to the training sample image;
for each of the training sample images,
obtaining at least one predicted region identifier probability map output by the first branch network layer, at least one predicted landmark identifier probability map output by the second branch network layer, and at least one predicted plane identifier probability map output by the third branch network layer, respectively, by inputting the training sample image into an initial neural network model;
determining a value of a target loss function based on the at least one predicted region identifier probability, the at least one predicted landmark identifier probability map, the at least one predicted plane identifier probability map, the at least one region identifier gold standard image, the at least one landmark identifier gold standard image, and the at least one plane identifier gold standard image; and
obtaining a trained neural network model by adjusting parameters of the initial neural network model based on the value of the target loss function corresponding to each of the training sample images.
16 . The system of claim 15 , wherein for each of the training sample images, the determining a value of a target loss function based on the at least one predicted region identifier probability, the at least one predicted landmark identifier probability map, the at least one predicted plane identifier probability map, the at least one region identifier gold standard image, the at least one landmark identifier gold standard image, and the at least one plane identifier gold standard image includes:
determining a value of a first loss function based on the at least one predicted region identifier probability map and the at least one region identifier gold standard image;
determining a value of a second loss function based on the at least one predicted landmark identifier probability map and the at least one landmark identifier gold standard image;
determining a value of a third loss function based on the at least one predicted plane identifier probability map and the at least one plane identifier gold standard image; and
determining the value of the target loss function by performing weighted summation on the value of the first loss function, the value of the second loss function, and the value of the third loss function.
17 . The system of claim 1 , wherein the image includes a magnetic resonance imaging (MRI) image.
18 . A method for brain identifier localization, implemented on a computing device having at least one processor and at least one storage device, the method comprising:
obtaining an image of a brain;
determining at least one region identifier probability map of the brain, at least one landmark identifier probability map of the brain, and at least one plane identifier probability map of the brain based on the image and a neural network model, wherein a value of each pixel point in the at least one region identifier probability map represents a probability value that the pixel point belongs to at least one region identifier of the brain, a value of each pixel point in the at least one landmark identifier probability map represents a probability value that the pixel point belongs to at least one landmark identifier of the brain, and a value of each pixel point in the at least one plane identifier probability map represents a probability value that the pixel point belongs to at least one plane identifier of the brain;
determining a segmentation result of a cerebral cortex of the brain, the at least one landmark identifier of the brain, and the at least one plane identifier of the brain, respectively, based on the at least one region identifier probability map, the at least one landmark identifier probability map, and the at least one plane identifier probability map;
constructing a target coordinate system based on the at least one landmark identifier and the at least one plane identifier; and
determining at least one landmark of the cerebral cortex based on at least one of the segmentation result of the cerebral cortex, the target coordinate system, and the at least one landmark identifier.
19 . The method of claim 18 , wherein:
the neural network model is a multi-task model including a shared network layer and three branch network layers, and the three branch network layers include a first branch network layer, a second branch network layer and a third branch network layer;
the first branch network layer is configured to segment the brain, and output the at least one region identifier probability map;
the second branch network layer is configured to locate the at least one landmark identifier of the brain, and output the at least one landmark identifier probability map; and
the third branch network layer is configured to locate the at least one plane identifier of the brain, and output the at least one plane identifier probability map.
20 . A non-transitory computer-readable medium, comprising executable instructions that, when executed by at least one processor, direct the at least one processor to perform a method, the method comprising:
obtaining an image of a brain;
determining at least one region identifier probability map of the brain, at least one landmark identifier probability map of the brain, and at least one plane identifier probability map of the brain based on the image and a neural network model, wherein a value of each pixel point in the at least one region identifier probability map represents a probability value that the pixel point belongs to at least one region identifier of the brain, a value of each pixel point in the at least one landmark identifier probability map represents a probability value that the pixel point belongs to at least one landmark identifier of the brain, and a value of each pixel point in the at least one plane identifier probability map represents a probability value that the pixel point belongs to at least one plane identifier of the brain;
determining a segmentation result of a cerebral cortex of the brain, the at least one landmark identifier of the brain, and the at least one plane identifier of the brain, respectively, based on the at least one region identifier probability map, the at least one landmark identifier probability map, and the at least one plane identifier probability map;
constructing a target coordinate system based on the at least one landmark identifier and the at least one plane identifier; and
determining at least one landmark of the cerebral cortex based on at least one of the segmentation result of the cerebral cortex, the target coordinate system, and the at least one landmark identifier.