Research On Adrenal CT Image Segmentation

Visual sense is the most direct way to obtain information. Image is an important carrier of information that brings people the rich visual effect. Natural image records the beautiful moment of objective world, remote sensing image presents the full physiognomy of the earth, medical image reveals the mystery of human life, images is a representation of objective object that can be found everywhere in our daily life. With the emergence of multi-core technology and the expansion of the memory, digital image processing technology is widely used. One of those technologies is image segmentation technique that is the process to segment image into different kinds of regions and to extract the interested target. So far, image segmentation technique has been widely employed in all areas such as medical science, military, remote sensing, artificial intelligence, product detection, computer vision, etc.With the continuous development of modern science and technology and the increasingly higher requirement to the accuracy of clinical diagnosis, medical imaging technology has developed rapidly, which has played a significant role in clinical medicine. All kinds of medical images not only enable doctor to see the change in form of internal tissue and organs, but also help to make an evaluation of the function of visceral organ. At present, all kinds of medical imaging technology, like ultrasound, X-ray, CT, MRI, nuclear medicine, have become an indispensable tool in clinical medicine and medical research. There are different respective applications between these different imaging methods. Ultrasound has the advantage of being non-intrusive, rapid in imaging process and high sensitivity. The imaging process of X-ray is simple, from which images can be directly produced by radiating fluorescent screen. X-ray image has the advantage of high spatial resolution. Currently, CT is one of the most widely used medical imaging equipments in clinical diagnosis. CT image belongs to internal slice image which is clear and has high density resolution. The advantages of MRI are non-radiative, multi-parameter imaging, rich contrast of tissue, and sensitive to early lesions, etc. With the help of nuclear medical imaging technique, doctor can not only observe the organ shapes, but also learn about their metabolism. The selective imaging and dynamic observation are the best advantages of nuclear medical imaging.Medical image segmentation is an important field in image segmentation. It refers to the process that carrying out mathematical calculations on origin medical image in order to extract the interested anatomical structure. The segmented result should be as similar to the real anatomical structure as possible. Medical image segmentation is a critical step in clinical auxiliary diagnosis, surgical planning and simulation, stereotactic radiotherapy and quantitative measurement, etc. Therefore, it is of great significance in clinical research. Different from natural image, medical image presents the complexity from image content and imaging conditions. Medical image has the large individual difference of human anatomy while its quality is easily affected by the offset field effect, the partial volume effect and all kinds of noise, which makes medical image segmentation a challenging work. Besides, the existing image segmentation algorithms still could not meet the requirements of clinical use. Hence, medical image segmentation technology has always been the hot topic of scientific research.With the spring up of medical image analysis, the research on medical image segmentation algorithms has been highly valued. So far, there are hundreds of image segmentation algorithms proposed by researchers. However, most of segmentation algorithms are aimed, to a large extent, at certain specific problems. There is still no universal theoretical guidance. Moreover, there is no standard answer to the classification of image segmentation algorithms. In this article, we classify the current segmentation algorithms as the bottom-up and top-down methods according to the design of algorithms. The bottom-up methods include threshold, region-based segmentation, clustering, markov random field, etc. The top-down methods contain active contour model (also known as snake), active shape model and level set method.The objects to be segmented in medical images are generally important organs or local lesion, such as brain tissue, liver, prostate and various tumor, etc. The adrenal glands are vital endocrine organs for releasing hormones to regulate body’s metabolism of sugar and salt and make cardiac muscle contract and elevate blood pressure. At present, the clinical diagnosis of cause of high blood pressure is still difficult. Doctors put forward a method that measuring the adrenal volume and configuration to analyze the cause of high blood pressure. To make the study proceed, it is necessary to segment the adrenal in CT image. The adrenal CT segmentation is valuable for volume calculation, anatomical location. However, the manual segmentation is difficult and time-consuming. To research the methods of adrenal CT segmentation is of great significance in clinic. Nowadays, the automatic segmentation of adrenal has not been effectively resolved.In adrenal CT segmentation, two factors should be considered.First, in CT image, the CT value of adrenal is similar to its adjacent tissue. The boundary of adrenal is not obvious. Thus, just using the CT value is hard to distinguish between adrenal and adjacent tissue. It is necessary to introduce more expressive features, like higher dimensional image features. Actually, higher dimensional characteristics of medical image have been successfully employed in other organs segmentation, such as prostate CT segmentation, liver ultrasound image segmentation and brain tumor segmentation, etc.On the other hand, adrenal shape has a strong statistical regularity. The priori shape information can provide important guidance for segmenting the adrenal. How to add the prior information in segmentation process is the key to improve the accuracy of adrenal segmentation result. Active Shape Model (ASM) is a kind of deformable model based on the statistical information on training set. Due to the convenience of embedding prior shape constraints, it has been widely employed in the segmentation of lungs, heart and so on. In this article, first, we have done a review about the popular algorithms of medical image segmentation. Then aiming at the characteristics of adrenal CT image and the effectiveness of current algorithms to adrenal segmentation, we emphasize the SVM and ASM methods. Finally, we propose a novel method based on the classifier on the higher dimensional image features and shape constraints for adrenal CT image segmentation. The main achievements of this article are listed as follows:(1) It gives a thorough view of all kinds of image segmentation algorithms. According to the design of algorithms, it classifies the current segmentation algorithms. These methods are analyzed in terms of the basic concept, algorithm principle, application field, advantages and disadvantages.(2) Using3D haar-like feature to describe the difference of inside and outside of adrenal. A novel feature extraction method based on higher dimensional features is proposed to classify the image.(3) It gives a full introduction of support vector machine and active shape model.(4) Aiming at the characteristics of adrenal CT image, since using a single image segmentation method for adrenal segmentation is not enough, it proposes a combined method that is based on classification of higher dimensional features and active shape model. This method uses their complementary advantages to improve the accuracy of adrenal segmentation.