Tree structure center line extraction method based on three-dimensional flux model

Abstract

The invention particularly relates to a tree structure center line extraction method based on a three-dimensional flux model, and the method comprises the steps: constructing a data set, and randomly sampling an original image to generate a sub-image; using a 3D Tubular Flux model to calculate the spatial distance and direction information of a foreground point, and generating the position information coding flux feature of the foreground point relative to the center line to serve as supervision information; performing feature learning by using a three-dimensional convolutional neural network to generate structural information flux features, enhancing channel exploration by using CBAM, and reducing the calculation amount by using a CSP module; performing adaptive center line prediction by using a lightweight U-Net network, and generating a dual-channel probability graph; processing the two-channel probability graph through a space weighted average strategy, and generating a finer center line; and solving the weighted mean square error loss of the three-dimensional convolutional neural network and the Dice loss of the lightweight U-Net network, carrying out summation, and then carrying out joint training. The end-to-end tree structure center line extraction method is realized, so that tree structure center line extraction is accurately and efficiently completed.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a method for extracting a centerline of a tree structure based on a three-dimensional flux model. Background technique [0002] In recent years, digital morphological reconstruction of tree-like structures such as neurons has become a research hotspot in the field of neuroscience. Reconstructing (tracing) tree-like structures from microscopic images is crucial for establishing the connectivity of brain circuits, conducting quantitative studies, and advancing the understanding of brain mechanisms. Therefore, many methods for automatic neuron reconstruction have been proposed in the past ten years, and the neuron reconstruction problem can be fundamentally formulated as centerline extraction, diameter estimation, and topological connectivity. Among them, extracting the centerline (skeleton) is a key part of the application of tubular structures, which provides a concise representati...

AI Technical Summary

Problems solved by technology

[0003] However, images of tree-like structures such as neurons have problems such as complex morphological structure, poor image quality, large noise, and broken texture. Causes errors in results such as incomplete centerline extraction, missing branches, premature termination and topology in closed branches

It makes it difficult for the existing tree-structure centerline extraction method to achieve high-efficiency and high-precision tree-structure centerline extraction

Images