Image detection method and device for aneurysm, computer equipment and storage medium

Abstract

The embodiment of the invention discloses an aneurysm image detection method and device, computer equipment and a storage medium, and the method comprises the steps: carrying out the three-dimensional image block extraction of a target head three-dimensional image through employing a preset sliding window and a preset sliding step length configuration, so as to serve as a target three-dimensional image block, the sliding step length, corresponding to the target direction, in the sliding step length configuration is smaller than the length, corresponding to the target direction, of the sliding window; inputting each target three-dimensional image block into a preset target prediction model, and performing three-dimensional image block aneurysm prediction, voxel aneurysm prediction and fusion processing to obtain an aneurysm mask block; and performing three-dimensional mask image splicing on the aneurysm mask blocks according to the target head three-dimensional image to obtain an aneurysm three-dimensional mask image. Therefore, completeness of the blood vessel is guaranteed as much as possible, confusion of features of the blood vessel truncation and features of the aneurysm is avoided, and false positive is reduced.

Description

technical field [0001] The present invention relates to the technical field of digital medicine, and in particular, to an image detection method, device, computer equipment and storage medium of an aneurysm. Background technique [0002] Intracranial aneurysms are a common and fatal disease, and rupture of the aneurysm can lead to subarachnoid hemorrhage, resulting in high mortality. Early detection of intracranial aneurysms facilitates its measurement, assessment of rupture risk, and clinical treatment of unruptured aneurysms. [0003] Most of the current image-based automatic detection methods for aneurysm use deep learning, but these methods have high false positives, with an average of 5-10 false positives per examination. The deep learning model is weak in distinguishing aneurysm and other arterial abnormalities, for example, it is weak in distinguishing blood vessel branches and blood vessel dilatation, and it is easy to misdiagnose normal arterial parts as aneurysms,...

AI Technical Summary

Problems solved by technology

[0003] At present, most of the automatic aneurysm detection methods based on images are realized by deep learning, but the false positives of these methods are high, with an average of 5-10 detection results being false positives for each inspection

The deep learning model is weak in distinguishing between aneurysms and other arterial abnormalities. For example, it is weak in distinguishing blood vessel branches and blood vessel expansion, and it is easy to misdiagnose normal arteries as aneurysms, resulting in high false positives.

Currently, the aneurysm prediction is performed by dividing 3D image blocks from the original 3D image. The blood vessels at the segmentation point of the two 3D image blocks are incomplete, which leads to confusion between the features of the cutoff part of the blood vessel and the features of the aneurysm, resulting in high false positives.

Patent Publication No. CN111369574A proposes a thoracic organ segmentation method and device. The scheme adopted in this patent is to directly determine the segmentation data as the segmentation result when the classification result is greater than the preset threshold value of the thoracic organ to be segmented. The method in the patent is to use the classification result as a judgment condition, so that false positives in the prediction process cannot be suppressed

The patent with the patent publication number CN113130078A proposes a method for predicting intracranial aneurysm occlusion. The scheme adopted in this patent is a weighted addition of the probability of including preset features and the probability of aneurysm occlusion. The method in this patent adopts The probabilities of different prediction targets are weighted and added, so that false positives in the prediction process cannot be suppressed

Images