Research On Segmentation Of Aortic Dissection Images

Automatic segmentation of aortic dissection images is the key to assisting in the diagnosis of aortic dissection.The segmentation of aortic dissection uses manual or semiautomatic methods to process CTA image data.These methods not only rely on experienced clinicians but are also time consuming and labor intensive.Therefore,research on more efficient aortic dissection image segmentation algorithms can help doctors diagnose aortic dissection.In this paper,a systematic study on the automatic segmentation of aortic dissection images is carried out to assist doctors in the rapid diagnosis of aortic dissection.The specific work and main results are as follows:(1)The CTA aortic dissection image database is established,and the CTA aortic dissection images are marked under the guidance of professional doctors.A variety of filtering algorithms are used to preprocess the CTA aortic dissection images,and the original images are enhanced using offset flipping,displacement,stretching,elastic deformation and other methods.(2)A two-stage aortic dissection image segmentation algorithm based on deep learning is proposed.The algorithm uses a two-stage training strategy to divide the segmentation task into two stages.In the first stage,the complete aorta is extracted by making full use of the image sequence spatial information to generate VOI(volume of interest).VOI determination not only ensures the integrity of key information but also achieves the effect of removing the image background.In the second stage,the model uses the VOI of the first stage as the input to accurately segment the pixels that are difficult to identify or are misidentified in the first stage to improve the segmentation accuracy of the aortic dissection.In the test,we use three different methods to determine the final output through voting to reduce the error generated during the upsampling process.(3)A deep learning method for aortic CTA image segmentation combined with an attention mechanism is proposed.Based on increasing the attention mechanism,considering the applicability of the loss function and the dataset,a hybrid loss function combining DICE(the Dice similarity coefficient)and IOU(intersection over union)is designed.The algorithm is validated and evaluated on the aortic dissection image database.The DICE coefficient is 91.43%,and the IOU coefficient is 84.21%,which verifies the effectiveness of the method.