Image contouring using spiking neural networks
Disclosed herein are methods and system for training artificial intelligence models configured to execute image segmentation techniques. The methods and system describe a server that receives a first image including a set of pixels depicting multiple objects. The server also receives a second image having a second set of pixels depicting the same set of objects. The server then analyzes the pixels from the first and second images. When a difference between at least one visual attribute of a pixel within the second image and a corresponding pixel within the first image satisfies a predetermined threshold, it will be encoded as spikes to send to the model, the model will be trained using supervised STDP rule by revising weights associated with the nodes within the AI model where the node corresponds to the pixels within the first and/or the second image.
1. A method comprising:
executing, by a computer, an artificial intelligence model utilizing a spiking neural network to segment one or more objects represented by a first set of pixels in a first image, the spiking neural network comprising a set of nodes corresponding to the first set of pixels, each node having a synaptic weight, wherein at least a first portion of nodes within the set of nodes represents input nodes corresponding to the first set of pixels and a second portion of the nodes within the set of nodes represents output nodes associated with a likelihood of each respective pixel corresponding to at least one object within the one or more objects;
comparing, by the computer, a grayscale attribute of each pixel within the first set of pixels with a corresponding pixel within a second set of pixels of a second image of the one or more objects;
iteratively training, by the computer using a supervised spiked-timing dependent plasticity method, the spiking neural network by:
in response to a difference between grayscale attributes of a first pixel within the first set of pixels with a second pixel within the second set of pixels satisfying a predetermined threshold:
a node corresponding to the first pixel transmits a spike, and
revising a synaptic weight associated with the first or the second pixel with each iteration; and
upon receiving a third image, executing, by the computer, the artificial intelligence model utilizing a trained spiking neural network to segment the one or more objects in the third image.
2. The method of claim 1 , wherein the computer revises the synaptic weight using the following formula:
Δ w=X trace ( T target −T actual )
wherein Δw represents a required weight change, wherein X trace represents a trace voltage from pre-node for post-node, wherein T target represents a ground truth value, and wherein T actual represents an actual spike output during training.
3. The method of claim 2 , wherein the computer depresses the synaptic weight when T target is lower than T actual .
4. The method of claim 2 , wherein the computer potentiates the synaptic weight when the T target is bigger than T actual .
5. The method of claim 1 , wherein the computer extracts at least one of the first image or the second image from a media element.
6. The method of claim 5 , wherein the media element is a four-dimensional computed tomography (CT) scan image.
7. The method of claim 6 , wherein the computer extract the at least one of the first image or the second image from the media element based on a predetermined time period.
8. The method of claim 7 , wherein the predetermined time period is received and revisable by a user operating a computing device that displays results of execution of the artificial intelligence model.
9. The method of claim 5 , wherein the third image is extracted from the media element.
10. The method of claim 1 , wherein the third image represents a neighboring phase of a four dimensional CT scan image.
11. A system comprising:
a server comprising a processor and a non-transitory computer-readable medium containing instructions that when executed by the processor causes the processor to perform operations comprising:
execute an artificial intelligence model utilizing a spiking neural network to segment one or more objects represented by a first set of pixels in a first three-dimensional image, the spiking neural network comprising a set of nodes corresponding to the first set of pixels, each node having a synaptic weight, wherein at least a first portion of nodes within the set of nodes represents input nodes corresponding to the first set of pixels and a second portion of the nodes within the set of nodes corresponds to a likelihood of each respective pixel corresponding to at least one object within the one or more objects;
compare a grayscale attribute of each pixel within the first set of pixels with a corresponding pixel within a second set of pixels of a second image of the one or more objects;
iteratively train using a supervised spiked-timing dependent plasticity method, the spiking neural network by:
in response to a difference between grayscale attributes of a first pixel within the first set of pixels with a second pixel within the second set of pixels satisfying a predetermined threshold:
a node corresponding to the first pixel transmits a spike having a synaptic weight, and
revising the synaptic weight associated with the first or the second pixel with each iteration; and
upon receiving a third image, execute the artificial intelligence model utilizing a trained spiking neural network to identify the one or more objects in the third image.
12. The system of claim 11 , wherein the server revises the synaptic weight using the following formula:
Δ w=X trace ( T target −T actual )
wherein Δw represents a required weight change, wherein X trace represents a trace voltage from pre-node for post-node, wherein T target represents a ground truth value, and wherein T actual represents an actual spike output during training.
13. The system of claim 11 , wherein the server depresses the synaptic weight when T target is lower than T actual .
14. The system of claim 11 , wherein the server potentiates the synaptic weight when the T target is bigger than T actual .
15. The system of claim 11 , wherein the server extract at least one of the first image or the second image from a media element.
16. The system of claim 15 , wherein the media element is a four-dimensional computed tomography (CT) scan image.
17. The system of claim 16 , wherein the server extract the at least one of the first image or the second image from the media element based on a predetermined time period.
18. The system of claim 17 , wherein the predetermined time period is received and revisable by a user operating a computing device that displays results of execution of the artificial intelligence model.
19. The system of claim 15 , wherein the third image is extracted from the media element.
20. The system of claim 11 , wherein the third image represents a neighboring phase of a four dimensional CT scan image.