Abstract
Improving the active surface model by using optimization of energy functions for the segmentation of 3D images
Recognizing the exact boundary and surface of objects in 2D or 3D images is considered one of the most important and complex issues in the field of image processing, which has many applications in the fields of machine vision, including It has object tracking, surface reconstruction and especially medical image processing. In the meantime, due to the increasing progress of medical science and its need for automatic and non-invasive diagnosis of all kinds of diseases and medical failures, the processing of two or three-dimensional medical images, especially three-dimensional images, is of particular importance because of the comprehensive and useful information they provide to the doctor. For this reason, various methods of segmentation of 3D images have been presented, which can generally be divided into three categories: structural, statistical and combined methods. Among these methods, parametric deformable models, which are among the structural methods, are of high importance and application. The basis of these models is based on the deformation of an initial surface due to the application of internal energy, which is responsible for the integrity and flexibility of the surface, and external energy, which is responsible for the movement of the initial surface towards the desired surface. Due to the proper performance of these models, more and more improvements are being made in this field.
In this thesis, the method of discrete active surfaces, which is one of the methods of parametric deformable models that has been presented recently and has shown very favorable results compared to the usual three-dimensional segmentation methods, is considered as the basic algorithm. In the proposed method, it has been tried to achieve a new and optimal method for segmentation of 3D images by improving the different steps of this algorithm and adding steps to complete and improve the results of the algorithm. For this purpose, the first improvement is about determining the appropriate initial surface, which is used here from an estimate obtained through the discrete active contour model, from the three-dimensional object as the initial surface. In the next part, in order to improve the internal and external energies, which have an important effect on the final result and the performance of the algorithm, the curvature integral is used as the internal energy, and the two local dependence functions of the phase and image gradient obtained from the boundary extraction by Violet are used as the external energy. Then, in order to correct the defects that occurred as a result of the algorithm, due to the presence of neighboring and similar areas, the linear search method was used. Finally, sampling and updating the triangular network based on the neighborhood was used to ensure the convergence of the algorithm, to extract areas with high curvature and to improve the accuracy of the extracted surface.
To check the performance of the proposed method in this thesis, this method is used as two algorithms on four categories of images, the image of a fake 3D star and CT scan images of brain, lung and liver have been applied and presented in the form of several image categories in the convergence stage as well as in the stopped stage in the iteration of the algorithm with a higher convergence speed and a table showing the computational load and speed of the proposed algorithms compared to the discrete active surfaces model. The evaluation of the results shows that the proposed method has a very favorable performance in terms of accuracy in extracting the final surface and areas with high curvature, as well as in terms of computational load and speed of convergence compared to the discrete active surface model.
Introduction
Image processing and analysis can be considered as a practical and technical structure to investigate, Analysis and extraction of information from defined images. Image segmentation is one of the most important and practical steps of image processing, which is used in many applications such as machine vision problems 1, feature extraction 2, object tracking 3, surface reconstruction 4, computer aided diagnosis 5, medical image processing 6 and many other applications.. In the meantime, medical image processing is extremely important considering that it is one of the most important tools for diagnosis, investigation and treatment of diseases for doctors and is widely used in the field of creating 2D, 3D and 4D images of the body and anatomical and physiological studies, and mainly imaging is done non-invasively without causing problems for the patient.
The most common and important methods used today for non-invasive medical imaging The following methods are used:
1 Machine Vision
2 Feature Extraction
3 Object Tracking
4 Surface Reconstruction
5 Computer Aided Diagnosis (CAD). A single photon3
-Magnetic resonance imaging4
-X-ray imaging5
-Computer radiography6
One ??of the most important steps in the analysis and processing of medical images is the segmentation of these images, which means defining the boundaries of one or more structures and anatomical parts of the body, because This work simplifies the processing and extraction of image information and also makes many parts of the image meaningful. In real medical images, due to various structural factors such as the anatomical structural complexity of the body, the proximity and overlapping of some organs, as well as external factors such as the presence of noise, low resolution of the images, blurring of the image due to the movement of the patient or distortions caused by the imaging device, the captured images have a lot of complexity for segmentation. Therefore, many algorithms may have problems in accurately and completely recognizing the borders and surfaces in medical images, and this problem has become one of the basic challenges in the segmentation of medical images.
Meanwhile, considering that 3D images contain valuable information of anatomical structures and body tissues such as volume, shape, size, location, presence of abnormalities and deformation of organs, segmentation of 3D images is increasingly used in the analysis of medical images. has For this reason, it is very important to provide methods and models to increase the accuracy and accuracy of 3D segmentation results. Computed Tomography (SPECT)
4 Magnetic Resonance Imaging (MRI)
5 X-Ray Tomography
6 Computed Tomography Scan (CT Scan)
1-2- Summary of the problem
In addition, 3D images, especially in medical science, especially because of having more information than 2D images and the possibility of easier processing and analysis for the user, the need to provide more accurate methods in this field is undeniable. On the other hand, segmentation of 3D images is one of the basic steps in processing such images. So far, many and various methods have been presented in the form of three categories of structural methods 1, statistical methods 2 and combined methods 3 to segment such images. In structural methods, the basis of the algorithm is based on the structural information of the image. While in statistical methods, the basis of the work is statistical and mathematical analysis of data. Also, combined methods try to use the properties of the previous two methods.
Although important steps have been taken in the general scheme of 3D image segmentation, there are still many difficult challenges for this problem, especially for medical images. For example, in most cases, the resulting image has noise and low contrast. On the other hand, the effect of the patient's movement and the relative volume in the imaging process can easily damage the image quality by blurring the edges of the tissue.