The Research On HIFU Ultrasound Images Segmentation

With the increasing pressure of modern life,people suffer from genital system diseases like uterine fibroid and prostate cancer.These diseases not only cause a decline in the quality of life but also the infertility issue.As a non-invasive and clinically wide-used treatment that kills cancer through heating effect,the High Intensity Focus Ultrasound(HIFU)has advantages of surgery needless,harmless to patients and without any influence on fertility.However,this approach faces the problem of how accurate,rapid,and automatic positioning of the lesion area during surgical procedures.In conventional clinical operations,the doctor needs to compare the MRI images with the ultrasound images of the same body position to confirm the position of the tumor.The procedure for one image of one body position costs 3-4 minutes.The entire surgical procedure generally requires the determination of the lesion area for dozens of ultrasound images of the position,and the time cost for locating the tumor alone is as high as 2-3 hours.The long period of surgery increased the workload of the doctor,leading to fatigue doctors and patients,is not conducive to medical safety.By using computer-aided techniques in the surgical procedure,the operation time can be greatly shortened and the working burden on doctors is reduced,so as to effectively avoid the medical safety problems caused by the exhaustion of doctors and patients.In HIFU,computer-aided localization of lesions based on computer vision,ultrasound image segmentation is the key technology to locate the lesion.Ultrasound images with low resolution,poor contrast,artifacts and other issues severely limited the development of ultrasonic image segmentation technology.Nowadays the process of mainstream ultrasonic image segmentation algorithm is complicated and long time-consuming,resulting in the difficulty to meet the clinical requirement.Aiming at HIFU ultrasonic image segmentation,a series of ultrasonic image segmentation algorithms are proposed in this paper,which improves the segmentation efficiency of ultrasonic images and achieves accurate segmentation result.The main contributions of these works are summarized as follows.(1)Ultrasound image segmentation based on Incorporate Spatial Information into Finite Rayleigh Mixture ModelThe dissertation proposed a model that incorporates the spatial information into Rayleigh Mixture Model for uterine ultrasound image segmentation.The Finite Mixture Model(FMM)is a common way for image classification.The experiments showed that this algorithm significantly reduces the influence of speckle noise and improves the accuracy of localization.The algorithm requires a small amount of competation,which consumes less computation time and the segmentation time of a single ultrasound image is less than 1 second,which fully meets the clinical needs.(2)Ultrasound image segmentation based on Bounded Rayleigh Mixture ModelThe dissertation proposed a model that incorporates the spatial information into Bounded Rayleigh Mixture Model(BRMM)for uterine ultrasound image segmentation.The BRMM adopts multiple Rayleigh distributions to fit the ultrasound image histogram more accurate.Meanwhile,the spatial information is incorporated into BRMM to reduce the speckle noise influence.The experiments show the model can further improve the accuracy of segmentation results.Its segmentation time of a single ultrasound image is less than 2 seconds,which fully meets the clinical needs.(3)Ultrasound image segmentation based on Weighted Finite Rayleigh Mixture ModelThe dissertation proposed a Weight Rayleigh Mixture Model(WRMM)that incorporates the weight spatial information into RMM for prostate ultrasound image segmentation.The model introduces saliency map for spatial information incorporation to reveal the importance degree of each pixel.The experimental results show accurate boundary and more detail in the segmentation that is benefit for tissue location in HIFU treatment.For ultrasonic image segmentation algorithm based on the WRMM model,a single image of the significant feature of the calculation time is less than 0.5 seconds,a single ultrasound image segmentation time within 2.5 seconds,fully meet the clinical needs.(4)Ultrasound image segmentation based on Improved Active Shape ModelThe dissertation proposed an Improved Active Shape Model that combines the prostate shape information and ultrasound image properties for prostate ultrasound image segmentation.The experiments show that the proposed model achieves accurate and rapid prostate segmentation effectively and reaches the level of the state of the art.