Research And Application Of Ventricular Segmentation Algorithm Based On MRI

Cardiovascular disease is the most fatal disease in the world.With the rapid development of medical imaging technology,the diagnostic procedures of cardiovascular disease have been facilitated.As a non-invasive imaging method,the short axis magnetic resonance image(MRI)are widely used in clinical diagnosis.In the diagnosis of cardiovascular disease,clinicians have to calculate cardiac function index to analyze cardiac function quantitatively.As a prerequisite,segmentation of cardiac MRI is an essential step.However,manual segmentation is not only time-consuming,but also prone to subjective errors.Therefore,there is an urgent demand for the efficient and accurate automatic segmentation method of cardiac MRI.Based on the public cardiac MRI dataset,the research on left ventricle,right ventricle and myocardium is implemented as follows: firstly,3D cardiac MRIs are transformed into 2D data by slicing.After that,resizing operation is performed to improve the efficiency of neural network training.Then rotation operation is implemented to augment data,avoiding over fitting.Finally,the gray value is normalized to reduce the calculation cost while maintaining the gray value difference.After the data preprocessing,the fully convolutional network(FCN)is modified to perform segmentation.Then the shortage of modified FCN is summarized and the dilated block adversarial network(DBAN)is proposed which consists of a segmentor and a discriminator.Within the segmentor which can generate segmentation probability maps,the dilated block is proposed to capture and aggregate multi-scale features.The discriminator adopting the fully convolutional scheme can differentiate the segmentation probability map from the Ground truth at the pixel level.Moreover,the discriminator can produce confidence probability maps helping the segmentor refine segmentation results through back propagation.The test results of DBAN are uploaded to the public dataset evaluation platform to assess the segmentation performance.The results show that the segmentation performance of DBAN can be ranked in the top 10 in terms of all metrics.Moreover,the segmentation results can be ranked top 3 in terms of seven metrics(totally12 metrics).Then,the heart function index based on the segmentation results of DBAN is evaluated,and the results showed that DBAN behave similar to medical experts.Therefore,it is reasonable to believe that DBAN is a potential candidate for automatic cardiac MRI segmentation.Based on the model of DBAN,an automatic cardiac MRI segmentation system is established.The system includes the functions of data import and saving,automatic segmentation and cardiac MRI viewing which can automatically and accurately segment cardiac MRIs.In addition,the system also includes the functions of extracting texture features and generating gray histogram,which help clinicians to check the lesions area more effectively and improve the diagnosis efficiency.