Linear structure extraction device, method, program, and learned model

A linear structure extraction method that extracts element points which constitute a linear structure from an image, the method comprising:

• by using a learning model implemented by a processor that is learned to receive an input of the image and output, as a prediction result, one or more element points which constitute the linear structure from the image, receiving the input of the image to the learning model;
• generating a feature map representing a feature amount of the image by performing convolution processing on the input image by using a first processing module;
• and dividing the feature map into a plurality of units including regions having a predetermined size in a grid pattern and calculating, for each unit, a shift amount from a unit center point to the element point of the linear structure closest to the unit center point by using a second processing module, wherein the second processing module disposes anchors, which are one or more reference shape regions having predetermined shape and size, in each of the units, and performs, for each unit, the convolution processing by using the feature amount at a position of the unit to calculate the shift amount for moving an anchor center point of the anchor to a closest point, which is the element point of the linear structure closest to the anchor center point, and a score for determining whether or not the linear structure is present in the anchors.