An Automatic Detection System of Cerebral Hemorrhage Based on Improved Unet

Abstract

The invention discloses an improved Unet-based automatic detection system for cerebral hemorrhage in CT images, which includes an image preprocessing module for obtaining brain CT images, and performing cutting and scaling processing on brain CT images after extracting brain parenchyma; The hemorrhage area detection module uses the improved Unet network to mark the position of the cerebral hemorrhage area on the brain CT image extracted from the brain parenchyma; the improved Unet network includes the RCSP convolution module, the CBL4 convolution module, the feature pyramid attention mechanism module, and the multi-scale A feature skip connection module and an output module; a data analysis module for estimating the total volume of hemorrhage and generating a three-dimensional image of the brain hemorrhage area. The invention realizes the automatic detection of cerebral hemorrhage, the estimation of the total volume of the hemorrhage area and three-dimensional imaging, effectively reduces the subjective error of manual segmentation and the workload of doctors, provides effective data support for clinical decision-making, and realizes Unet to detect cerebral hemorrhage , using more contextual semantic information in CT images, the detection accuracy is higher.

Description

technical field [0001] The invention relates to the field of hemorrhage recognition on CT slices, in particular to an improved Unet-based automatic detection system for cerebral hemorrhage in CT images. Background technique [0002] Spontaneous cerebral hemorrhage is a neurological emergency with high morbidity and is the main cause of stroke. It accounts for 9%-13% of stroke in high-income countries, and the proportion of stroke in China has reached 25%. In patients with cerebral hemorrhage, the location of hemorrhage and the volume of hematoma play an important role in the prognosis and decision-making of diagnosis and treatment of patients with cerebral hemorrhage. CT imaging examination can directly display and observe the lesions. [0003] The traditional manual marker segmentation is the gold standard for obtaining the hematoma volume in the hemorrhage area from CT images. This method is not only time-consuming and laborious, but also has a large Individual measurers ...

AI Technical Summary

Problems solved by technology

The hemorrhage area and non-hemorrhage area of ​​the patient's brain CT image have a high similarity in grayscale features, and the experimental samples are relatively small. Due to the simple structure of the Unet network, the feature extraction ability of the hemorrhage area is insufficient, resulting in its model training. Overfitting phenomenon, and the segmentation results are relatively poor

Images