Detect and Identify Aneurysms Based on Adjusted 3D Attention UNet

Early diagnosis and treatment of cerebral aneurysms are important for reducing the risk of aneurysm rupture. Fast and accurate detection of aneurysms on blood vessels is a key step in diagnosis of aneurysm. To date, a large number of deep learning algorithms, especially the UNet network, have been developed for detection of aneurysms. However, when the amount of data for training is small, it is difficult to obtain a reliable deep learning network to effectively identify aneurysms. In order to address this issue and improve the accuracy of aneurysm detection, here we proposed to combine the deep learning approach with specially designed preprocessing and postprocessing algorithm. We first determined the rough locations of the aneurysms based on the features on the vascular skeleton before aneurysms segmentation with deep learning network, i.e. 3D Attention UNet in this work, thus reducing the missed detection rate of the UNet network. We could obtain the shape and texture related to the aneurysm. Then we used the random forest algorithm to implement the feature classification model to find out the false aneurysms incorrectly detected by the U-Net network. The experimental results show that our method can accurately identify aneurysms in the case of small data sets.