Research On Target Tracking Methods Of Contrast-enhanced Ultrasound Lesion Area

In the medical field,doctors only rely on single frame image information for lesion detection,but lack comprehensive and effective information in the time dimension.The Region of Interest(ROI)is a part of the lesion area,also known as a suspected lesion area.The tracking of the object area can effectively locate the lesion area.In order to diagnose the lesion area more accurately,the sonographer manually select ROI in each frame of the video sequence,and perform global observation according to the information of the entire time dimension.If we can use computer means to assist doctors to tracking object changes,not only save doctor's time,but also improve the efficiency of medical diagnosis.Therefore,the research on target tracking methods of contrast-enhanced ultrasound lesion area was established to assist doctors in making scientific judgments on the disease and reduce the impact of subjective judgments.The main contents of this thesis include three parts: pretreatment of Contrast-enhanced Ultrasound image,fusion differential optical flow and deep learning feature for object tracking and the target object tracking model of cross-modal data migration.Contrast-enhanced ultrasound image preprocessing include image denoising and image enhancement.In this thesis,a novel anisotropic diffusion model based on edge enhancement(EEAD)is applied for the speckle noise in contrast-enhanced ultrasound image without causing detail loss.The MSRCR algorithm is improved for the dark image caused by contrast media has not arrived yet.It is proposed that an MSRCP method based on simple color balance to suppress the features of uninterested regions.On one hand,it enriches the effective information,on the other hand,it strengthens the need for image interpretation and target tracking in next step.It is proposed that the target tracking of differential optical flow and fusion deep learning feature.For the phenomenon of particle drift and degradation that occurs in the process of particle tracking.Firstly the optical flow method is used to perform global detection on the previous frame of image,and the weights of the particles are calculated and redistributed by the optical flow component.Then deep learning instead of the traditional method in feature extraction.Finally candidate particle region is input into convolutional neural network.Experiments show that the algorithm can improve the particle drift effectively and reduce tracking error as well as improve the tracking accuracy.The target tracking model for cross-modal data migration is proposed.We take advantage of the commonality between different modes of medical image data and use ultrasound image as bridge.By migrating the trained natural image model to the contrast-enhanced image,the model is finally applied to the feature extraction of particle filter frame for object tracking.Experiments show that the trained model can extract the feature of the contrast-enhanced ultrasound image effectively and have good accuracy.In addition,the idea of cross-modal data migration can be used not only in the field of tracking,but also in the identification of benign and malignant images.