Method and system for generating a centerline for an object, and computer readable medium

A system for generating a centerline for an object, comprising:

• an interface configured to receive an image containing the object, wherein the image is acquired by an imaging device;
• and
• a processor configured to:
• generate the centerline of the object by tracing a sequence of patches with a virtual agent, including for each patch other than the initial patch:
• determine a current patch based on the position and action of the virtual agent at a previous patch;
• determine a policy function and a value function based on the current patch using a trained learning network, which comprises an encoder followed by a first learning network and a second learning network, wherein the learning network is trained by maximizing a cumulative reward and minimizing an auxiliary loss of detecting at least a bifurcation in a trajectory of the object, wherein at least one of the first learning network or the second learning network comprises a multi-layer perceptron layer or a recurrent neural network;
• and
• determine the action of the virtual agent at the current patch;
• and
• a display configured to display the centerline of the object.

A non-transitory computer readable medium storing instructions that, when executed by a processor, perform a method for generating a centerline for an object, the method comprising:

• receiving an image containing the object, wherein the image is acquired by an imaging device;
• generating the centerline of the object by tracing a sequence of patches with a virtual agent, by the processor;
• including, for each patch other than the initial patch:
• determining a current patch based on the position and action of the virtual agent at a previous patch;
• determining a policy function and a value function based on the current patch using a trained learning network, which comprises an encoder followed by a first learning network and a second learning network, wherein the learning network is trained by maximizing a cumulative reward and minimizing an auxiliary loss of detecting at least a bifurcation in a trajectory of the object, wherein at least one of the first learning network or the second learning network comprises a multi-layer perceptron layer or a recurrent neural network;
• and
• determining the action of the virtual agent at the current patch;
• and
• outputting the centerline of the object for display.