Diffusion Magnetic Resonance Imaging Synthesis Of Cardiac Fiber Based On Deep Learning

Studying the structure of myocardium is of great fundamental and clinical importance for understanding the causes and adopting early diagnosis of various cardiovascular diseases.At present,diffusion magnetic resonance imaging(d MRI)is the main method for non-destructive detection of myocardial fibrous tissue structure,but this method is particularly sensitive to heart beats,making the collected images have motion blur effects,and it is impossible to obtain the entire cardiac cycle of d MRI images.In addition,due to ethical limitations,the ex-vivo heart data is difficult to obtain,so the cardiac d MRI data is very limited.With the emergence of big data and deep learning,the use of large amounts of medical imaging data to explore the causes and early diagnosis of major diseases has become a trend.Therefore,In this paper,we investigated the several deep learning models to synthesize a large number of cardiac d MRI data that are similar to the real acquisitions.The main research works of this thesis are as follows:(1)We used the DCGAN network to synthesize a large amount of cardiac d MRI data.The synthesized d MRI image is very similar to the real acquired image visually,and it can accurately extract myocardial fibers from the synthesized image.(2)To promote further the cardiac d MRI synthesis quality,d MRI image was firstly synthesized using analytical model,and then we design a d MRI-GAN model.This model was modified based on pix2 pix model,it includes the SE module,perceptual loss and Wasserstein GAN loss.The experimental results illustrate that the d MRI-GAN model performs better than the pix2 pix model,and can synthesize realistic DW images,which can also reflect the microstructure of actual tissues.(3)Due to motion effect in the in vivo cardiac d MRI data,the acquired in vivo d MRI data experience the signal loss.This paper uses a model to complete the in vivo cardiac DW image.Such model consists of an edge generation model and a content completion model.The model can restore the outline of the image,complete the structural information of the irregularly-shaped missing regions,and restore the fiber direction and diffusion characteristics of the missing regions.