Systems and methods for multi-label segmentation of cardiac computed tomography and angiography images using deep neural networks

A method comprising:

• training a deep neural network to map 3D CCTA images to 3D multi-label segmentation maps;
• receiving a first 3D CCTA image;
• mapping the first 3D CCTA image to a 3D multi-label segmentation map using the trained deep neural network;
• determining a plurality of 1D parametric curves for a coronary artery included in the first 3D CCTA image using the 3D multi-label segmentation map;
• and
• determining a location of a lesion in a branch of a coronary tree using the plurality of 1D parametric curves, wherein determining the location of the lesion in the branch of the coronary tree using the plurality of 1D parametric curves comprises determining start and stop points of the lesion in the branch of the coronary tree based on the plurality of 1D parametric curves,
• wherein determining the plurality of 1D parametric curves for the coronary artery comprises determining values for each of a plurality of pre-defined parameters for each point along a centerline of the branch of the coronary tree, the plurality of pre-defined parameters include lipid core thickness, vessel wall thickness, lumen radius, and lumen tortuosity.

A cardiac computed tomography and angiography (CCTA) imaging system comprising:

• an x-ray radiation source;
• an x-ray detector array;
• a memory storing a trained deep neural network and instructions;
• and
• a processor communicably coupled to the x-ray radiation source, the x-ray detector array, and the memory, and when executing the instructions, configured to:
• acquire a 3D CCTA image of a coronary tree using the x-ray radiation source and the x-ray detector array;
• map the 3D CCTA image to a multi-label segmentation map with the trained deep neural network;
• generate a plurality of 1D parametric curves for a branch of the coronary tree using the multi-label segmentation map;
• and
• determine a location of a lesion in the branch of the coronary tree using the plurality of 1D parametric curves,
• wherein the processor is configured to:
• generate the plurality of 1D parametric curves for the branch of the coronary tree using the multi-label segmentation map by determining values for each of a plurality of pre-defined parameters along a centerline of the branch of the coronary tree, wherein the plurality of pre-defined parameters are derived from the multi-label segmentation map, wherein the plurality of pre-defined parameters include one or more of lipid core thickness, vessel wall thickness, lumen radius, and lumen tortuosity;
• and
• determine the location of the lesion in the branch of the coronary tree using the plurality of 1D parametric curves by:
• comparing the plurality of 1D parametric curves against one or more pre-defined thresholds;
• and
• determining start and stop points of the lesion in the branch of the coronary tree based on one or more of the plurality of 1D parametric curves crossing one or more of the pre-defined thresholds.