The Research Of Parametric Multi-objective Optimization In Graph-based Segmentation For Breast Ultrasound Images

As a special women disease, breast cancer severely threatens the health and lives of women. However, the cause of it is still unknown for human, and early detection and treatment is the key to reduce the death rate. As an important method for breast cancer detection, ultrasonic imaging detection depends on clinicians ’ observation and diagnosis, which makes the diagnostic result kind of subjective and uncertain. Therefore, computer-aided diagnosis(CAD) has been proposed with the development of computer techniques and image processing techniques. As an important part of CAD system based on breast ultrasound image, image segmentation is the connecting link between the preceding and the following, and the s egmentation result directly determines the accuracy and reliability of diagnostic result. Therefore, the research of techniques for breast ultrasound image segmentation has very important significance. However, various ultrasonic artifacts(high speckle noise, low contrast, weak boundary, intensity inhomogeneity, and so on) make accurate segmentation of ultrasound image a difficult task.In the last few years, graph-based segmentation has become a research hotspot due to the simple structure and solid theories. RGB algorithm, which is an advanced graph-based image segmentation algorithm, can accurately segment the breast ultrasound image. However, there are two significant parameters which should be manually set in RGB algorithm. To obtain good segmentation results, different parameters may be set for different breast ultrasound images, which greatly constrains the clinical application of RGB algorithm. Therefore, to make RGB algorithm be able to automatically set good parameters according to different images, particle swarm optimization(PSO) algorithm is adopted to multi-objectively optimize the two parameters of RGB algorithm, and an algorithm called multi-objectively-optimized RGB(MOORGB) algorithm is proposed in this paper.The main steps of MOORGB algorithm are: cropping TCI subimage, speckle reduction, contrast enhancement, homogeneity improvement, RGB segmentation, objective function calculation, PSO optimization, boundary smooth ing, and so on. The main innovation points are:(1) contrast enhancement a nd homogeneity improvement are added into the image preprocessing part due to various ultrasonic artifacts(high speckle noise, low contrast, intensity inhomogeneity, and so on);(2) two novel and targeted optimal objective functions are proposed;(3) base d on the idea of aggregating functions, two novel methods to handle multiple optimal objective functions are proposed;(4) the parameter setting and end condition has been improved to improve the performance and efficiency;(5) the image postprocessing par t is added to smooth the boundary of segmentation result. To validate the feasibility and effectiveness of MOORGB algorithm, a relatively large dataset(including 100 clinical breast tumor ultrasound images) has been adopted to perform the experiment, and the experimental results has been compared with those of other three image segmentation algorithms proposed in recent years. The experimental results show that the proposed MOORGB algorithm has significant improvement, better avoiding over and under segmen tation, and obtains more accurate segmentation results for breast tumor ultrasound images.