Research On The Detection Method Of Cerebral Hemorrhage Based On CT Image

Cerebral hemorrhage is a kind of primary non-traumatic intraparenchymal hemorrhage.When the blood vessels inside the brain rupture under pressure,it will cause cerebral hemorrhage.At present,most of the diagnosis of cerebral hemorrhage is based on CT images,and then doctors make judgments on the CT images of cerebral hemorrhage.However,in emergency situations,doctors cannot determine whether a patient has cerebral hemorrhage symptoms in a timely and accurate manner through CT images,thus missing the patient’s optimal treatment time.Therefore,it is imperative to find a method to automatically identify the location of cerebral hemorrhage from brain CT images.For this reason,in the field of clinical medicine and image detection,many researches have been conducted on the automatic detection and recognition methods of cerebral hemorrhage.Traditional methods for automatically detecting CT images of cerebral hemorrhage often face various problems such as low accuracy,cumbersome methods,and slow speed,and the final detection results are average.With the development of deep learning,people began to use deep learning to detect related lesions in medical images.Compared with traditional detection methods,detection methods based on deep learning can improve the accuracy and speed of lesion detection in most cases.Therefore,the use of deep learning to detect related lesions has gradually become a better solution in the field of medical image detection.The main work of this paper is to study the cerebral hemorrhage detection algorithm based on brain CT images,which is mainly divided into two parts,namely the segmentation of brain parenchyma and the detection of cerebral hemorrhage.For brain substance segmentation,this article first describes a traditional segmentation algorithm based on linear spatial filters,and then proposes a U-Net brain substance segmentation algorithm based on residual network.This algorithm modifies the network structure and adds The depth of the network.This article elaborates the improvement details of U-Net network structure in detail,and compares and analyzes the segmentation algorithms based on linear spatial filter,U-Net and improved U-Net network respectively.For the detection of cerebral hemorrhage,this article briefly describes the current popular target detection algorithm,and selects Mask-RCNN as the main algorithm for cerebral hemorrhage detection.This article describes several improved algorithms for Mask-RCNN feature extraction network Res Net101,and proposes another improved method,which improves the running speed of the algorithm without reducing the network’s receptive field.Then it analyzes the shortcomings of the FPN network,and proposes an improved FPN network structure based on the original FPN network.This network structure integrates more levels of features and improves the detection accuracy of the algorithm.Experiments prove that the brain parenchymal segmentation method and cerebral hemorrhage detection method proposed in this paper have high accuracy and efficiency.