Sequence pulmonary nodule image segmentation method based on superpixels and density clustering

Abstract

The invention discloses a sequence pulmonary nodule image segmentation method based on superpixels and density clustering. The method comprises the following steps that preprocessing is conducted by combining average intensity projection (AIP) of three-dimensional characteristics of a CT image with multi-scale dot enhancement; over-segmentation is conducted on the image according to a hexagonal clustering and morphological optimizing superpixel sequence image segmentation algorithm (HMSLIC); adaptive weighting coefficients are constructed in distance computing of superpixel blocks, precise positioning of pulmonary nodules is achieved to obtain a follow-up clustering start block, and a follow-up accurate clustering threshold value is obtained by fitting the distances and detecting changes of the slope; and by adopting the fast DBSCAN superpixel sequence image clustering algorithm of which the strategy of only clustering the pulmonary nodules and threshold adaption are optimized, a sequence pulmonary nodule mask is obtained, and finally sequence pulmonary nodule images of a lung CT are obtained. Various types of sequence pulmonary nodule images can be more quickly, integrally and accurately obtained through segmentation, and a basis is laid for follow-up treatment and diagnosis analysis.

Description

technical field [0001] The invention relates to the segmentation of pulmonary nodule sequence images, in particular to a sequential pulmonary nodule image segmentation method based on superpixels and density clustering. Background technique [0002] Fast and accurate segmentation of pulmonary nodule sequence images is the basis for subsequent processing and diagnostic analysis. In the previous research on sequential pulmonary nodule image segmentation algorithms, the segmentation of solid nodules is very effective, but the following problems arise: without reducing the segmentation accuracy, the sequential pulmonary nodule images cannot be efficiently segmented ; There is a large difference in the gray value of the cavity inside the cavitary nodule and the surrounding area, and it is easy to regard the cavity as a part of the lung parenchyma, resulting in incomplete segmentation of the nodule area; the gray value of the vascular adhesion type nodule and the blood vessel The...

AI Technical Summary

Problems solved by technology

In the previous research on sequential pulmonary nodule image segmentation algorithms, the segmentation of solid nodules is very effective, but the following problems arise: without reducing the segmentation accuracy, the sequential pulmonary nodule images cannot be efficiently segmented ; There is a large difference in the gray value of the cavity inside the cavitary nodule and the surrounding area, and it is easy to regard the cavity as a part of the lung parenchyma, resulting in incomplete segmentation of the nodule area; the gray value of the vascular adhesion type nodule and the blood vessel The gray value is very close, it is easy to keep the blood vessel as part of the nodule, and cannot effectively separate the blood vessel from the nodule; there is a lot of noise, which leads to the segmentation of the pulmonary nodule and the noise and various tissue structures Therefore, a method that can not only achieve accurate segmentation of sequential pulmonary nodule images, but also meet the clinical requirements in terms of speed has important reference value for assisting physicians to make reliable diagnostic decisions

Images