Intracranial hemorrhage segmentation method fusing dense connection and attention mechanism

Abstract

The invention aims at solving the problems that an intracranial hemorrhage area is not clear in structure, artifacts exist, and other brain tissue and other noise cause great influences on a segmentation task. In order to improve the performance of intracranial hemorrhage segmentation, an intracranial hemorrhage segmentation method fusing dense connection and attention mechanisms is provided. Intensive connection blocks are introduced into an encoder part of a full convolutional network for intracranial hemorrhage feature extraction, but not all features extracted from an encoder can be used for segmentation, so that an attention mechanism fusing space and channel attention is fused into a network architecture, intracranial hemorrhage features are weighted in terms of space and channel, rich context relations are captured, and more accurate features are obtained. In addition, a Focal Tversky loss function is adopted to process segmentation of small-area intracranial hemorrhage. The segmentation performance is effectively improved, and accurate and rapid segmentation can be realized.

Description

technical field [0001] The invention relates to medical image processing and semantic segmentation, and specifically designs an intracranial hemorrhage segmentation method integrating dense connection and attention mechanism. Background technique [0002] Intracranial hemorrhage is a neurological disease caused by the rupture of blood vessels into the tissue and may extend the endothelial cells in the brain. For a long time, intracranial hemorrhage is the main cause of death and disability paralysis. After traumatic brain injury, if not diagnosed in time and treatment, most likely to induce higher mortality [1] . In the traditional diagnostic method, doctors manually estimate the area and size of bleeding. The whole process is very time-consuming, and the determination of the type and area of ​​bleeding depends on the accumulated experience of doctors. Human error is also the problem that leads to deviations in the diagnosis of intracranial hemorrhage. [0003] With the de...

AI Technical Summary

Problems solved by technology

It is used to solve various problems such as imaging artifacts in intracranial CT images, noise in brain tissues such as skulls, and relatively unclear structures in intracranial hemorrhage areas. Labeling and training, finely extracting the features of intracranial hemorrhage lesions by means of full convolutional network fusion of dense connection blocks and attention mechanism

For small intracranial hemorrhage areas, this method introduces the Focal Tversky loss function to solve the problem of insufficient segmentation accuracy for small area hemorrhage lesions, effectively improving the performance of intracranial hemorrhage segmentation

Images