Method and system for hemorrhage area segmentation in brain CT images based on semi-supervised learning

Abstract

The invention discloses a method for segmenting bleeding areas in brain CT images based on semi-supervised learning. The method includes a semi-supervised model training stage and a semi-supervised model-based bleeding area segmentation stage; the semi-supervised model training stage is used to train semi-supervised model; the bleeding area segmentation stage based on the semi-supervised model includes format conversion of the two-dimensional CT image sequence that needs to be segmented into the intracranial hemorrhage area, reconstructing the two-dimensional CT image into a three-dimensional space, and then using the supervoxel algorithm to divide the three-dimensional image into For supervoxels of similar size, features are extracted from each supervoxel as a sample, and finally the supervoxels are divided into foreground and background parts through a trained semi-supervised model based on the features. The present invention effectively improves the accuracy of bleeding area detection by introducing a semi-supervised learning algorithm and using supervoxels instead of pixels for calculation.

Description

technical field [0001] The present invention relates to the fields of machine learning and image processing, in particular to a method and system for segmenting hemorrhage areas in brain CT images based on semi-supervised learning. Background technique [0002] Intracranial hemorrhage (ICH) is one of the most serious acute cerebrovascular diseases, and it is also an important cause of acute neurological disorders such as hemiplegia. Therefore, for clinical treatment, early diagnosis of intracranial hemorrhage is of great significance. Compared with clinical manifestations, computed tomography (CT) scan and magnetic resonance imaging (MRI) scan, which can be detected without trauma, can more directly and accurately reflect the severity and evolution trend of intracranial hemorrhage. At the same time, because the cost of CT detection is much less than that of MRI detection, most patients will choose CT detection. Fresh hematomas usually appear as hyperluminous areas with ind...

AI Technical Summary

Problems solved by technology

First, most of these methods use very simple segmentation algorithms, such as clustering and thresholding, etc., and while these methods may perform well in natural image processing, they do not perform well in complex situations, such as bleeding regions overlapping brain tissue. These methods do not work well when the edges of the bleed or bleed are not sufficiently discriminative

Second, most of the existing algorithms are only suitable for processing two-dimensional images

However, CT imaging is a three-dimensional process, so a series of parallel scanning image frames will be generated, while the 2D segmentation algorithm will ignore some important inter-frame information

Images