Master thesis Atomic and Nuclear physics: Eesearch and development of spect and spect CT images segmentation software for automatic detection and extraction of brain tumors using ITK, VTK, QT

Chia sẻ: _ _ | Ngày: | Loại File: PDF | Số trang:85

Thêm vào BST

Báo xấu

15
lượt xem 2
download

Download Vui lòng tải xuống để xem tài liệu đầy đủ

In this thesis, the author focussed on providing an application framework to be used in brain segmentation - the process of tumor detection. This constitutes setting up software bundled into a Window machine which is easily distributable. Moreover, a segmentation algorithm was implemented and tested using the provided software. In order to test the bundled software by implementing specific brain segmentation from the literature, the author used CT scans freely available from the database of 108 Central Military Hospital.

Chủ đề:

Bình luận(0) Đăng nhập để gửi bình luận!

Lưu

Nội dung Text: Master thesis Atomic and Nuclear physics: Eesearch and development of spect and spect CT images segmentation software for automatic detection and extraction of brain tumors using ITK, VTK, QT

MINISTRY OF EDUCATION VIETNAM ACADEMY OF AND TRAINING SCIENCE AND TECHNOLOGY GRADUATE UNIVERSITY OF SCIENCE AND TECHNOLOGY ----------------------------- Ho Thi Thao RESEARCH AND DEVELOPMENT OF SPECT AND SPECT/CT IMAGES SEGMENTATION SOFTWARE FOR AUTOMATIC DETECTION AND EXTRACTION OF BRAIN TUMORS USING ITK, VTK, QT MASTER THESIS: ATOMIC AND NUCLEAR PHYSICS Hanoi - 2019
MINISTRY OF EDUCATION VIETNAM ACADEMY OF AND TRAINING SCIENCE AND TECHNOLOGY GRADUATE UNIVERSITY OF SCIENCE AND TECHNOLOGY ----------------------------- Ho Thi Thao RESEARCH AND DEVELOPMENT OF SPECT AND SPECT/CT IMAGES SEGMENTATION SOFTWARE FOR AUTOMATIC DETECTION AND EXTRACTION OF BRAIN TUMORS USING ITK, VTK, QT Major: Atomic and nuclear physics Code: 60440106 MASTER THESIS: ATOMIC AND NUCLEAR PHYSICS SUPERVISORS: Dr. Phan Viet Cuong MSc. Le Tuan Anh Hanoi - 2019 i
Confirmation This thesis was written on the basic of my research works carries out at Institute of Physics, Vietnam Academy of Science and Technology under the supervision of Dr. Phan Viet Cuong and MSc. Le Tuan Anh. All results of other authors that are used in this thesis are cited correctly. April 20, 2018 The author Ho Thi Thao ii
Acknowledgements I would like to express my gratitude to all people who have helped and inspired me during my study. This thesis would not have been possible without those supports from many people. First of all, I would like to thank my teachers, Dr. Phan Viet Cuong and MSc Le Tuan Anh, Research and Development Center for Radiation technology, Vietnam Atomic Energy Institute, for giving me the opportunity to do within their group, for their guidance, support and constant encouragement during the entire period of preparation of this thesis. They had often pointed out the incompleteness of my work and helped me to improve my understandings on each problem. They taught me a lot about nuclear physics, nuclear medicine, coding and all academic and non-academic matters. I have been extremely lucky to have supervisors who cared so much about my work, my study and who responded to my questions and queries so promptly. I have learned a lot of things from them, and more importantly they showed me that everything can be done, just keep hard working, keep big dream, keep big courage, keep going on, day by day . In addition, I would also like to thank all the members at Center for Nuclear physics, Institute of Physics, Vietnam Academy of Science and Technology, for providing the best possible environment for us to study and research. Finally, I would like to thank Mom, Dad, my brother, for their constant love and support. My sister has helped me to diminish the fact of being away from home by the long telephone calls spent laughing. I would also like to thank my friends, Luan, Ha, my sister, Tan,… made my time at VAST a lot more fun. To everybody else who accompanied me throughout my time as a student: Thank you! Ho Thi Thao iii
Abstract Digital Imaging and Communications in Medicine (DICOM) exists as a standard for handling, storing, printing, and transmitting information in medical imaging. The DICOM files include not only the information of images, but also contain a lot of medical-related information. Reading and Process an image in DICOM format is an important issue for further image processing and visualization. In the field of medical image processing, detection of brain tumor from computed tomography (CT), magnetic resonance (MRI), positron emission tomography (PET) or single-photon emission computed tomography (SPECT) is a difficult task due to complexity of the brain hence it is one of the top priority goals. In this thesis, the author describes a new method which combines four different steps including smoothing, Sobel edge detection, connected component and finally region growing algorithms for locating and extracting the various lesions in the brain. The computational algorithm was implemented by INMOFEVV a new software which combines Insight Toolkit (ITK) to process input image, Visualization Toolkit (VTK) to display and Qt software development framework to build user interface. The main function of software includes reading and displaying DICOM images as well as performing advanced image processing. It helps to improve quality and efficiency of the diseases diagnosis. The analysis results indicate that the proposed method automatically and efficiently detected the tumor region from brain medical images. It is very clear for physicians to separate the abnormal from the normal surrounding tissue to get a real identification of related area; improving quality and accuracy of diagnosis, which would help to increase success possibility in treatment by early detection of tumor as well as reducing surgical planning time. Key Words: DICOM, Image Processing, ITK, VTK, QT. ix
x Table of contents Acknowledgements ..................................................................................... iii Abstract ....................................................................................................... ix Table of contents ......................................................................................... x Index of figures .......................................................................................... xii List of acnonyms ....................................................................................... xvi INTRODUTION .......................................................................................... 1 CHAPER 1. OVERVIEW .......................................................................... 5 1.1. INTERACTION OF RADIATION WITH MATTER .............. 5 1.1.1. Interaction of photons with matter ..................................... 5 1.1.1.1. Types of photon interactions in matter .............................. 5 1.1.1.2. Attenuation of photons in matter ....................................... 6 1.1.2. Interaction of charged particles with matter ..................... 7 1.2. SINGLE PHOTON EMISSION COMPUTED TOMOGRAPHY AND COMPUTED TOMOGRAPHY ........................ 8 1.2.1. Single-photon Emission Computed Tomography .............. 8 1.2.1.1. Gamma camera .................................................................. 8 1.2.1.2. Single photon emission computed tomography (SPECT) 15 1.2.2. Computed tomography ....................................................... 18 1.2.3. Hybrid Imaging System: SPECT/CT............................... 23 CHAPTER 2. EXPERIMETAL AND METHODOLOGY .................. 27 2.1. DICOM..................................................................................... 27 2.1.1. DICOM image ........................................................................ 27 2.1.2. DICOM Information Model .................................................. 27 2.2. RECONSTRUCTION .............................................................. 28 2.2.1. Iterative Reconstruction Method .......................................... 29 2.2.2. Filtered Backprojection Method (FBP) ............................... 29 2.2.3. Filtering ................................................................................... 30 2.3. TECHNOLOGIES ................................................................... 30 2.3.1. Build System - CMake ........................................................ 30 2.3.2. Source Code Libraries ........................................................ 31 2.3.3. Database ............................................................................... 31 2.4. INITIAL RESULTS OF INMOFEVV SOFTWARE .............. 32 2.4.1. Fusion images ...................................................................... 32
xi 2.4.3. Surface and volume rendering ........................................... 34 2.4.4. Filters .................................................................................... 35 2.5.1. General description of proposed method.......................... 38 2.5.2. Preprocessing: Mean filter ................................................. 41 2.5.2.1. Factors affecting image quality CT, MRI scan................ 41 2.5.2.2. Mean filter ........................................................................ 42 2.5.3. Sobel edge detection ............................................................ 42 2.5.4. Segmentation ....................................................................... 43 2.5.4.1. Connected Component Labeling ...................................... 43 2.5.4.2. Region growing by Confidence Connected...................... 44 3.1. CHOOSING SUITABLE FILTER FILTER FOR PREPROCESSING STEP ................................................................ 47 3.2. CHOOSING SUITABLE METHOD FOR EDGE DETECTION STEP .......................................................................... 50 3.3. APPLYING PROPOSED METHOD FOR PHANTOM GAMEX 463....................................................................................... 50 3.3.1. Method and test results on phantom ................................... 51 3.3.2. Result evaluation ................................................................. 57 3.4. APPLYING PROPOSED METHOD FOR BRAIN IMAGES............................................................................................. 59 CONCLUSIONS ....................................................................................... 67 REFERENCES .......................................................................................... 69
xii Index of figures Figure 1.1. Predominant types of interaction for a range of incident photon energies and absorber atomic numbers. .............................................. 6 Figure 1.2. Attenuation.......................................................................................... 7 Figure 1.3. Penetrating and nonpenetraiting radiation. ......................................... 8 Figure 1.4. Components of a standard nuclear medicine imaging system. ............ 9 Figure 1.5. Collimator detail ................................................................................ 10 Figure 1.6. Scintillation crystal. A sodium iodide crystal ―doped‖ with a thallium impurity is used to convert gamma photons into light photons. ...... 11 Figure 1.7. Sodium iodide crystal scintillation detector. .................................... 13 Figure 1.8. Photomultiplier tube and its preamplifer and amplifer. .................... 13 Figure 1.9. The positioning algorithm improves image resolution. .................... 14 Figure 1.10. Small matrix. ................................................................................... 16 Figure 1.11. Storing image data in a matrix. ....................................................... 16 Figure 1.12. SPECT camera. ............................................................................... 17 Figure 1.13. Three-headed SPECT camera ......................................................... 17 Figure 1.14. Confgurations of two-headed SPECT camera. ............................... 17 Figure 1.15. (A.) Slices through the level of the heart from selected projection views are stacked to create a sinogram. (B) Complete sinogram. .... 19 Figure 1.16. Basic components of one type of CT scanner, containing a stationary detector ring and rotating inner X-ray tube. .................... 20 Figure 1.17. Rotate–stationary configuration. A rotating source and collimator generate a fan-shaped X-ray beam that is directed toward a stationary ring of detectors. ............................................................................... 20 Figure 1.18. Rotate–rotate configuration. The opposing source and detector rotate synchronously. ........................................................................ 20
xiii Figure 1.19. Multislice CT detector array composed of multiple rows of detectors placed side by side along the z-axis. ................................. 21 Figure 1.120. SPECT-CT. (a) Two-gantry system with CT system contained within one gantry and SPECT heads supported on a second gantry; (b) Single-gantry system with one gantry supporting both the SPECT camera heads and an X-ray tube and detector. ............................. 26 Figure 2.1. DICOM Information Object Definition (IOD) of a patient. ............. 28 Figure 2.2. Projection views of a liver are backprojected to create transaxial slices. ................................................................................................ 29 Figure 2.3. Star artifact and backprojection ―blur‖ artifact. ................................ 30 Figure 2.4. Interface of image processing software ............................................ 32 Figure 2.5. Fusion result of CT and SPECT image (a,b,c) and SPECT/CT (d) obtained from machine of 108 Central Military Hospital. ............... 33 This is the original fusion result of INMOFEVV software. The correct SPECT/CT image must be captured at the same time. DICOM images obtained from 2 devices must be the same size, recorded time,… Overcoming the disadvantages of fusionimages from any two devices: time, dose projection, shooting angle,… is a big challenge for image reconstrution algorithms. ................................. 34 Figure 2.6. Multiplanar reconstruction a, brain image; b, abdominal image ...... 34 Figure 2.7. 3D visualization results. .................................................................... 35 Figure 2.8. Some filters are often used. .............................................................. 36 Figure 2.9. The distribution of contour lines....................................................... 37 Figure 2.10. Flowchart of the proposed method ................................................. 39 Figure 2.11. Proposed image segmentation algorithm ........................................ 40 Figure 2.12. Sobel edge detection algorithm. ..................................................... 43 Figure 3.1. The graph shows the change of area by threshold. ........................... 57 Figure 3.2. Three regions are extracted corresponding to the selected seed points of Gamex phantom. .......................................................................... 58
xiv Figure 3.3. Raw input images.............................................................................. 60 Figure 3.4. Preprocessing mean filter ................................................................. 60 Figure 3.5. Sobel edge detection. ........................................................................ 60 Figure 3.6. Connected component without using Sobel. .................................... 62 Figure 3.7. Connected component using Sobel. .................................................. 62 Figure 3.8. Segmented images after using region growing. ................................ 63 Figure 3.9. The results of extracting large and small brain tumors use the proposed segmentation method on MRI images. ............................. 65 Figure 3.10. The results of extracting large and small brain tumors use the proposed fragmentation method on lung images.............................. 65
xv List of Tables Table 1.1. Hounsfield values of some tissues………………………………….17 Table 3.1. Results of MSE, PSNR values of 4 Noise filters: Mean, Median, Gauss and Bilateral…………………………………………………42 Table 3.2. Dependency of segmentation results on input threshold value with minsize= 20, 50, 100, 200. …………………………………………45 Table 3.3. Dependency of segmentation results on input threshold value with minsize= 20…………………………………………………………47 Table 3.4. Results of seed points, area of 3 round holes extracted within the threshold of 15. ……………………………………………………..53 Table 3.5. Results of evaluation of MSE and PSNR indicators for edge detection methods……………………………………………………………. 55 Table 3.6. Area of the extracted tumor…………………………………………58 Table 3.7. Results of brain tumor extraction (1, 2, 3, 4, 5) and lung tumor (6), tumor in liver (7) and extraction of interest area 8 of region growing method………………………………………………………………60
xvi List of acnonyms Acronym Meaning DICOM Digital Imaging and Communications in Medicine ITK Insight Segmentation and Registration Toolkit VTK Visualization Toolkit CT Computed tomography SPECT Single-photon emission computed tomography PET Positron Emission Tomography PACS Picture archiving and communication system MRI Magnetic resonance LEAP Low-energy all-purpose collimators PMT Photomultiplier tube ADC Analog-to-digital converter GBPS Gated Blood Pool Study MUGA Multigated Acquisition ART Algebraic reconstruction technique MLEM Maximum likelihood expectation maximization OSEM Ordered-subsets expectation maximization FBP Filtered Backprojection Method NEMA National Electrical Manufacturers' Association Qt Cute  VR Value representation VL Value length PMTs Photomultiplier tubes HU Hounsfield units
1 INTRODUTION Together with the development of the image diagnostics technology, nowadays most hospitals have been already equipped with a variety of digital imaging equipments, and, PACS - Picture archiving and transmission system has been established. In the field of the medical imaging, DICOM is the standard used for the storage and transmission of medical images which can provide the interface standards and protocols for the manufacturers and users of the medical imaging equipment. The interpretation of DICOM medical image files, the reading of medical image data, the display, and processing of image processing are very important. Most of medical imaging equipment such as Computed tomography (CT), Magnetic resonance (MR), Positron emission tomography (PET), Single photon emission computed tomography (SPECT) etc. have supported DICOM standard. Currently, numerous foreign software packages are available for medical image processing and analysis in Vietnam, such as eFilm, 3D-Doctor, DICOMWorks, BrainSuite etc. Difficulty in equipping and using such packages in Vietnam hospitals are their high price and proprietary technology of the manufacturer. There are many groups that have written about medical image processing software in Vietnam. However their application software, DROC, V-Doctor, BKDICOM etc. has a limited number of functionalities: Image processing and enhancement functions are still limited, so it is difficult to understand the detail anatomical structure of the patient; Most are not upgraded frequently; The ability to protect patient information has not yet been developed; They have not been applied in practice for communication between doctors and patients. From above facts, we have built a multi-purpose medical image processing application featuring enhancement, segmentation of multimodal images obtained from different equipment. In the frame of this dissertation, the author describes about our development of the software to read DICOM images of CT, MRI and SPECT.
2 Brain tumors are known to be one of the main diseases leading to human death in the world. A brain tumor varies according to its location, size, shape, and appearance. Early detection of brain tumor tends to be very challenging as normally there seem to be no clear symptoms from the beginning stages. Clearly visible from CT, MRI images, there is overlap between the boundaries of the tumor in surrounding and tissue, the edges can be obscured by the structure of the skull, resulting in a lot of contrast to the background. Therefore, it is difficult to distinguish the boundary between normal and abnormal tissues. Removing the tumor without affecting the surrounding tissues is a big challenge for the doctors [1]. The continuous need for enhanced and accurate automated brain segmentation and detection methods is an important part of computer-aided diagnosis. Other requirements include: fast processing time; high-level of automation avoiding the need for manual intervention; low cost, maintenance and support requirement including training; ability to shapes, sizes, and types. In this thesis, the author focussed on providing an application framework to be used in brain segmentation - the process of tumor detection. This constitutes setting up software bundled into a Window machine which is easily distributable. Moreover, a segmentation algorithm was implemented and tested using the provided software. In order to test the bundled software by implementing specific brain segmentation from the literature, the author used CT scans freely available from the database of 108 Central Military Hospital. Early detection and treatment can increase the rate of survival for patients. There are many approaches used in many researches to differentiate biological tissue edges of brain images. Biji et al. [2] proposed a technique to detect tumors from MR images using fuzzy clustering and minimum error thresholding. This method shows how this technique overcomes the problem of over-segmentation with watershed algorithm but the major drawback is the computational time required. Bhattacharyya et al. [3] have concluded that a set algorithms based on thresholding are a powerful tool for the detection of
3 brain tumor in MRI images. The method proposed by Anam Mustaqeem, Ali Javed, Tehseen Fatima [4] required a watershed algorithm for segmentation. The article goes into anatomical analysis of the brain and symptoms, damage caused by encephalopathy. The only downside is the over-segmentation leads to poor detection of significant areas with low contrast boundaries that commonly results in MRI brain images. M.C. Jobin Christ and R.M.S. Parvathi [5] have introduced the method of the brain tumor detection that integrates K Means clustering with a marker-controlled watershed algorithm and integrates Fuzzy C Means clustering with marker-controlled watershed algorithm separately for medical image segmentation. The drawback of K- means clustering is that it requires multiple loops. M. Masroor Ahmed et al. [6] proposed the method for the detection and extraction of brain tumor from MRI images using K-means Clustering. This method is very effective, that is proven to be less time consuming and achieves maximum lossless data compression. But this approach causes fake edges on the image. Up to now, there are many different algorithms that have been proposed and implemented. And each of the technique has its own advantage and disadvantage. However, there is no single approach that can generally solve the problem of segmentation for the large variety of image modalities existing today. Segmentation algorithms most effectively are obtained by customizing combinations of components carefully. Parameters of these components are adjusted for the characteristics of the imaging method used as input and for the features of the segmented anatomy [7]. Many researches, softwares in the world today focus on the segmentation of medical images such as Slicer3D, Osirix [8, 9],... are either manual or semi-manual, takes a lot of processing time. Tumor detection is a long and time-consuming process. Location determination, characterization of the tumor depends much on the experience and skill of the doctor. Most of these segmentation works are manually done by hand. Manual segments are often inaccurate. If manual adjustments bring good results, it is not practical
4 for large datasets. Thus, the location of tumor is needed to determine automatically. In this thesis, the author uses a region growing method for efficient segmentation with marking the region of interest (ROI) as well as the background in gray image. This process combines the basic approaches: smoothing, edge detection, and region growing segmentation. Here, the author proposed mean smoothing in order to reduce the noises in CT, MRI images. Sobel algorithm is used for image segmentation. It uses the connected component as well to set proper boundaries between adjacent regions. The texture feature is extracted using region growing method. Hence, it is easy to implement and provides more stable results than using individual methods. The Insight Toolkit is an open source cross-platform application toolkit widely used by researchers in the field of medical image processing. In this thesis, an environment containing the newest versions of ITK integrated together with Visualization Toolkit and Qt framework was prepared. Moreover, a detailed literature survey relating to brain segmentation was also carried out which resulted in developing a semi-automated region growing- based brain segmentation method implemented and tested using a database from 108 Central Military Hospital. The content of this dissertation includes four chapters: The first begins with introduction and motivation, the current literature and state of the art techniques for brain segmentation. Chapter 1 dives into the structure, basic principle, basic characteristics of SPECT and hybrid imaging system SPECT/CT. Chapter 2 describes algorithms, filters commonly used in SPECT, SPECT/CT image processing, and their advantages and disadvantages. These filters are discussed and they are realized by software programming. The segmentation of tumor using edge and region growing operations is discussed in detail. Chapter 3 discusses the structure of proposed approaches and results.
5 CHAPER 1. OVERVIEW 1.1. INTERACTION OF RADIATION WITH MATTER When radiation strikes matter, both the nature of the radiation and the composition of the matter affect what happens. The process begins with the transfer of radiation energy to the atoms and molecules, heating the matter or even modifying its structure. If all the energy of a bombarding particle or photon is transferred, the radiation will appear to have been stopped within the irradiated matter. Conversely, if the energy is not completely deposited in the matter, the remaining energy will emerge as though the matter were transparent or at least translucent. The thesis will introduce some of the physical phenomena that are involved as radiation interacts with matter, the interactions in matter of both photons (gamma rays and X-rays) and charged particles (alpha and beta particles). 1.1.1. Interaction of photons with matter As they pass through matter, photons interact with atoms. The type of interaction is a function of the energy of the photons and the atomic number (Z) of the elements composing the matter. 1.1.1.1. Types of photon interactions in matter In the practice of nuclear medicine, where gamma rays with energies between 50 and 550 keV are used, Compton scattering is the dominant type of interaction in materials with lower atomic numbers, such as human tissue (Z = 7.5). The photoelectric effect is the dominant type of interaction in materials with higher atomic numbers, such as lead (Z = 82). A third type of interaction of photons with matter, pair production, only occurs with very high photon energies (greater than 1020 keV) and is therefore not important in clinical
6 nuclear medicine. Figure 1.1 depicts the predominant types of interaction for various combinations of incident photons and absorber atomic numbers. Figure 1.1. Predominant types of interaction for a range of incident photon energies and absorber atomic numbers. 1.1.1.2. Attenuation of photons in matter As a result of the interactions between photons and matter, the intensity of the beam, that is, the number of photons remaining in the beam, decreases as the beam passes through matter (Figure 1.2). This loss of photons is called attenuation. Specifically, attenuation is the ratio of the intensity at the point where the beam exits the attenuator, I out, to the intensity it had where it entered, Iin. The attenuation is an exponential function of the thickness x of the attenuator in centimeters. This resembles the exponential manner in which radioactivity decays with time. Expressed symbolically:
7 Iout/Iin=e−µx (1.2) where μ, the linear attenuation coefficient. The linear attenuation coefficient is greater for dense tissue such as bone than for soft tissue such as fat. In general, the linear attenuation coefficient depends on both the energy of the photons and on the average atomic number (Z) and the thickness of the attenuator. The lower the energy of the photons or the greater the average atomic number or thickness of the attenuator, the greater the attenuation. Figure 1.2. Attenuation 1.1.2. Interaction of charged particles with matter Because of the strong electrical force between a charged particle and the atoms of an absorber, charged particles can be stopped by matter with relative ease. Compared with photons, they transfer a greater amount of energy in a shorter distance and come to rest more rapidly. For this reason, they are referred to as nonpenetrating radiation (Figure 1.3).