Detection Of Cerebral Aneurysms Based On Deep Learning

Intracranial aneurysm is a kind of cerebrovascular disease,once the aneurysm ruptures,if not timely help,it is likely to endanger the safety of the patient’s life.In the diagnosis of intracranial aneurysms,it is mainly to review the TOF-MRA type slice data to determine whether a brain tumor has occurred.Therefore,hospital radiologists need to spend a lot of time and energy on the diagnosis of radiography every day,and the diagnosis result is easily affected by the doctor’s own state and experience of radiography reading at that time.With the rapid development of deep learning,more and more researchers are paying attention to the use of deep learning-based object detection technology to automatically detect tumor targets in medical images.As an important part of medical diagnosis,detection technology is the key to a revolutionary breakthrough in the future.However,there are still many shortcomings in the current tumor detection methods,including :(1)small object detection has always been a difficulty in the field of vision.As there are large and small brain tumor targets,most of the existing detection methods have not been ideal for small object detection;(2)in the medical image data of continuous slices,most of the existing detection methods ignore the time sequence information between slices,and focus more on the extraction of image feature information.Although the deep feature information can well represent the image content,the time sequence information between the upper and lower sections is helpful to improve the accuracy of brain tumor detection;(3)in the medical image,the non-focus area is easy to interfere with the detection of the target object,resulting in false detection or missed detection.The existing detection methods cannot well identify the features of similar objects in the image,thus resulting in the decline of detection accuracy.Based on the problems mentioned above,this article proposes a new improved detection method for brain tumor detection in medical images.The main research work is summarized as follows:1.For the problems of different sizes of brain tumors,this article will use YOLO V3 as the basic network to solve.Because of the multi-scale fusion of YOLO V3,the multi-scale strategy performs very well in targets with large size variations.In addition,the detection speed of YOLO V3 algorithm is very fast,far faster than other detection algorithms,which also meets the needs of auxiliary diagnosis.2.For the problem of time series information extraction,this article will adopt LSTM model to solve it.After extracting rich and robust visual features by using the deep convolutional neural network in YOLO V3,the feature information is then passed into LSTM to construct the timing information between upper and lower continuous slices.The learning ability of the whole network is extended to time sequence,not limited to image feature learning.At the same time,the LSTM model also has a strong regression ability,which can infer the region position of the target object from the visual features,which is helpful for the subsequent detection to obtain more accurate object coordinate position.3.Finally,spatial attention mechanism model is integrated.The attentional mechanism is used to focus on the feature extraction of the brain tumor to reduce the interference of the blood vessel part to the detection.Because vascular parts and brain tumors are similar in many features,they need to be distinguished from each other.Spatial attention can increase the weight of brain tumor and enhance the detection performance.The experimental method proposed in this article was tested on the data set of the Union Hospital(Wuhan)and laboratory cooperation project.The experimental results verify the effectiveness of the proposed innovation points,in which the m AP evaluation index can reach 70.9%,which proves the feasibility of the idea proposed in this paper.In short,the research contents and ideas of this article have important research value for the detection of brain tumors in medical images.