Method and system for anatomical tree structure analysis
The present disclosure is directed to a computer-implemented method and system for anatomical tree structure analysis. The method includes receiving model inputs for a set of positions in an anatomical tree structure. The method further includes applying, by a processor, a set of encoders to the model inputs. Each encoder is configured to extract features from the model input at a corresponding position. The method also includes applying, by the processor, a tree structured network to the extracted features. The tree structured network has a plurality of nodes each connected to one or more of the encoders, and information propagates among the nodes of the tree structured network according to spatial constraints of the anatomical tree structure. The method additionally includes providing an output of the tree structured network as an analysis result of the anatomical tree structure analysis.
1. A computer-implemented method for an anatomical tree structure analysis, comprising:
receiving model inputs for a set of positions in an anatomical tree structure, wherein the anatomical tree structure includes at least one bifurcation point and a plurality of branches splitting from the at least one bifurcation point;
applying, by a processor, a set of encoders to the model inputs, wherein each encoder is configured to extract features from the model input at a corresponding position;
applying, by the processor, a tree structured network to the extracted features, wherein the tree structured network is a neural network that has a plurality of nodes constructed according to the anatomical tree structure and each node of the tree structured network is connected to one or more of the encoders, wherein information propagates among the nodes of the tree structured network according to spatial constraints of the anatomical tree structure; and
providing an output of the tree structured network as an analysis result of the anatomical tree structure analysis.
2. The method of claim 1 , wherein the anatomical tree structure is a blood vessel or an airway.
3. The method of claim 1 , wherein the set of positions include the at least one bifurcation point and at least one point in each branch.
4. The method of claim 1 , further comprising:
receiving an image of the anatomical tree structure acquired by an image acquisition device; and
deriving the model inputs at the set of positions in the anatomical tree structure from the image.
5. The method of claim 1 , wherein the encoders are selected from a convolutional neural network (CNN), a fully convolutional neural network (FCN), and a multi-layer perceptron (MLP).
6. The method of claim 1 , wherein the tree structured network is a recurrent neural network (RNN) that includes a plurality of RNN unit each corresponding to a node.
7. The method claim 6 , wherein the RNN units are selected from a long short-term memory (LSTM) and a gate recurrent unit (GRU).
8. The method of claim 1 , wherein the information propagates bi-directionally between a pair of nodes corresponding to two adjacent positions in a path of the anatomical tree structure.
9. The method of claim 1 , wherein the information propagates in a single direction between a pair of nodes, wherein the single direction is either from a distal side to a root between the pair of nodes, or from the root to the distal side between the pair of nodes.
10. The method of claim 1 , wherein the set of encoders and the tree structured network are trained jointly.
11. A system for performing an anatomical tree structure analysis, comprising:
an interface, configured to receive model inputs for a set of positions in an anatomical tree structure, wherein the anatomical tree structure includes at least one bifurcation point and a plurality of branches splitting from the at least one bifurcation point; and
a processor, configured to:
apply a set of encoders to the model inputs, wherein each encoder is configured to extract features from the model input at a corresponding position;
apply a tree structured network to the extracted features, wherein the tree structured network is a neural network that has a plurality of nodes constructed according to the anatomical tree structure and each node of the tree structured network is connected to one or more of the encoders, wherein information propagates among the nodes of the tree structured network according to spatial constraints of the anatomical tree structure; and
provide an output of the tree structured network as an analysis result of the anatomical tree structure analysis.
12. The system of claim 11 , wherein the anatomical tree structure is a blood vessel or an airway.
13. The system of claim 11 , wherein the set of positions include the at least one bifurcation point and at least one point in each branch.
14. The system of claim 11 , wherein the interface is further configured to receive an image of the anatomical tree structure acquired by an image acquisition device, and wherein the processor is further configured to derive the model inputs at the set of positions in the anatomical tree structure from the image.
15. The system of claim 11 , wherein the encoders are selected from a convolutional neural network (CNN), a fully convolutional neural network (FCN), and a multi-layer perceptron (MLP).
16. The system of claim 11 , wherein the tree structured network is a recurrent neural network (RNN) that includes a plurality of RNN unit each corresponding to a node.
17. The system of claim 16 , wherein the RNN units are selected from a long short-term memory (LSTM) and a gate recurrent unit (GRU).
18. The system of claim 11 , wherein the information propagates bi-directionally between a pair of nodes corresponding to two adjacent positions in a path of the anatomical tree structure.
19. The system of claim 11 , wherein the information propagates in a single direction between a pair of nodes, wherein the single direction is from a distal side to a root between the pair of nodes, or from the root to the distal side between the pair of nodes.
20. A non-transitory computer readable medium having instructions stored thereon, the instructions, when executed by a processor, perform a method for an anatomical tree structure analysis, the method comprising:
receiving model inputs for a set of positions in an anatomical tree structure, wherein the anatomical tree structure includes at least one bifurcation point and a plurality of branches splitting from the at least one bifurcation point;
applying a set of encoders to the model inputs, wherein each encoder is configured to extract features from the model input at a corresponding position;
applying a tree structured network to the extracted features, wherein the tree structured network is a neural network that has a plurality of nodes constructed according to the anatomical tree structure and each node of the tree structured network is connected to one or more of the encoders, wherein information propagates among the nodes of the tree structured network according to spatial constraints of the anatomical tree structure; and
providing an output of the tree structured network as an analysis result of the anatomical tree structure analysis.