Sayısal hücre görüntülerinin kodlanması ve nicel analizi
Coding and quantitative analysis of the digital cell images
- Tez No: 39123
- Danışmanlar: Y.DOÇ.DR. MUHİTTİN GÖKMEN
- Tez Türü: Yüksek Lisans
- Konular: Elektrik ve Elektronik Mühendisliği, Patoloji, Electrical and Electronics Engineering, Pathology
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 1993
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Belirtilmemiş.
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 86
Özet
ÖZET Bu çalışmada sayısal tıp görüntülerinden olan hücre görüntülerinin analizi ve kodlanması konuları ele alınmıştır. Burada kullanılan hücre görüntüleri Marmara Üniversitesi Tıp Fakültesi Patoloji bölümünden temin edilmiş hastalıklı beyin dokusuna ait mikroskobik görüntülerdir. Sayısal hücre görüntülerinin fazlalığı, bunların etkin bir şekilde depolanmasını gerektirmiştir. Tıp görüntülerinde detayların önemli olması nedeniyle özgün görüntüye ulaşma imkanı veren kayıpsız sıkıştırma yöntemleri incelenmiş ve bu görüntülere Duruk (static) Huffman kodlama yöntemi uygulanmıştır. Elde edilen sıkıştırma oranlarının yükseltilmesine çalışılmış, bunun için uzamsal ilintiyi azaltan yöntemler denenmiştir. Hücre görüntülerinin içerdikleri bilgilerin hastalıkların teşhis ve tedavisinde büyük rol oynaması üzerine bunların otomatik olarak analiz edilmesine çalışılmıştır. Analizlerin otomatik hale getirilmesi ile daha kısa sürede, daha doğru sonuçlar elde edilmesi amaçlanmıştır. Hücre görüntülerinde bulunan çekirdek ve çekirdekçiklerin sayısı ve alanlarının, uzmanlar tarafından önemli görülmesi, analizin hedefini belirlemiştir. Analizin ilk ve en zor adımı olan çekirdek ve çekirdekçik sınırlarının saptanmasında var olan kenar saptama yöntemleri uygulanmıştır. Bu yöntemlerden daha iyi sonuçlar elde edebilmek için görüntüleri, bu yöntemlere daha uygun hale getirecek işlemler uygulanmıştır. Çekirdek ve çekirdekçiklerin alanlarının hesabı, hiyerarşik yapının çıkarılması sınırların saptanmasını takip eden işlemlerdir. Analiz sonucu elde edilen bilgilerin bir dosyada saklanarak, daha sonraki aşamalarda analize gerek duyulmadan kullanılabilmesi sağlamıştır. Bu dosyalardaki bilgiler kullanılarak görüntülerin simgesel gösterimleri elde edilmiştir. Bu simgesel gösterimlerde çekirdek ve çekirdekçikler önce karelerle, daha sonra çekirdek ve çekirdekçik şekillerinin dairelere benzemesi üzerine dairelerle ifade edilmişlerdir. Bu çalışmanın amacı, sayısal tıp görüntülerin depolanma sorununa çözüm getirmek ve bu görüntülerin analizlerinin otomatik olarak gerçekleştirilmesini sağlayarak bunun önemini ortaya koymaktır. Çalışma boyunca, bunları gerçekleştirirken karşılaşılan zorlukları belirtilmiş ve bu zorlukların giderilmesi için çözümler aranmıştır. v
Özet (Çeviri)
SUMM ARY CODING AND QUANTITATIVE ANALYSIS OF THE DIGITAL CELL IMAGES Digital image processing, the manipulation of images by computer, rel- atively recent development in terms of man's visual stumuli. in it's short history, it has been applied to pratically every type of imagery, with varying degrees of success. it is a vast umbrella under which fail diverse aspect of optics, electronic, mathematics, photography, and computer technology. Digital image processing can be grouped into these areas: image en- hancement, restoration, coding, segmentation, and analysis. in image en- hancement,images either are processed for human viewers. in image restora¬ tion, an image has been degraded in some manner and the objective is to reduce ör eliminate the eifect of degradation. The objective in image coding is to represent an image with as few bits as possible, preserving a certain level of image quality and intelligibility. image segmentation is the divi- sion of an image into different regions, each having certain properties. The applications of image segmentation are numerous. image segmentation has been used in biomedical area such as in the identification lung diseases, in automated classification of white blood cells, in detection of cancerous cells. The objective of image analysis is to symbolically represent the contents of the image. image analysis is distinguished from other types of images pro¬ cessing, in that the ultimate product of an image analysis system is usually numerical output rather than an image. Digital image processing is closely tied to human vision, which is öne of the most important means by which humans percieve the world. As a result, new applications continue to be found and existing applications continue to expand in such diverse areas, such as communication, defense, consumer electronics, medicine, robotics, and geophysics. This study is concerned with öne of these areas, medicine. With the continuing development in medical imaging demand for image storage and image analysis is increasing rapidly. This study was done on these subjects, on the microscobic celi images,getting from the brain tissue. The images are obtained from the Pataloghy of Marmara Medicine Faculty. viThe amount of data associated with visual information is so large that its storage would require enormous storage capacity. Storage of this data requires large capacity, which could be very expensive. For the purpose of archiving, it is mandatory that the effîciency of representation of the image is optimal. in image data compression, the goal is to represent the image with as few bits as possible while preserving the level of the quality and intelligibility required for given application. image compression is the art and science of fmding efficient representa¬ tion for images. in general compression techniques can be divided in two majör groups. l.Reversible, lossless compression. 2.1rreversible, lossly compression. The reversible compression is the öne which encodes an image by remov- ing ör at least reducing the redundancy in the image data in such a way that original data can be exactly reconstructed from the encoded image data. The irreversible compression is the öne whose encoding process will not keep ali of the original information, but reconstructed image retains the original image quality to a larger degree. The compression ratio, which can be defined as the ratio of the aver- age bits per symbol before compression, to average bits per symbol after compression, is considerably sınailer for the reversible techniques than that achievable by the irreversible techniques. The attainable compression ratio depends the image type, spatial resolution and histogram characteristics of the image. Lossless data compression methods like DCT (discrete cosine transform), vector quantization approach compression ratios of 10 ör 20 to 1. Ali these methods introduce some noise and artifacts, in the sense that reconstruc- tion is not identical to original. Since the human visual system is imperfect, these reconstructions may be very good, indistinguishable from the original ör at least adaquate. While indistinguishable may be adaquate for a great deal of applications, but not for medical, artistic, defense and security images. As the details can be more important in these applications, the reversible techniques are used. That is the answer of why lossless compression techniques are used in this study for medical images. On the other hand, image data does have a quality in which the neighbouring pixels are highly correlated. If these interdependencies are removed as much as possible, better compression ra¬ tios could be obtained. viiin this study, the lossless compression methods, such as arithmetic cod- ing, static Huffman coding, dynamic Huffman coding, Lempel-Ziv coding, Lempel-Ziv-Welch coding, and run-length coding are examined. Arithmetic coding maps a sequence of data symbols to an encoded se- quence in such a way that the original data can be recovered from the encoded sequence. Artihmetic coding algorithm consists of data model, statistics estimation and arithmetic coder. The model determines the cur- rent event to be encoded and its context. The estîmation method computes the relative frequency distribution used for each context (provided by the model). The encoder accepts the events to be encoded and generates the code sequence. The encoding and decoding algorithms perform arithmetic operations on the encoded sequence. The coding algorithm is recursive, and on each recursion, the algorithm successively partitions an interval of the number line between O and 1. Huffman coding reduces the number of bits used to represent frequent symbols and increases the number of bits used for infrequent symbols. Static Huffman coding requires that a table of probabilities must be available be- fore beginning to compress data. The cömpressor prescans the input data and finds the symbol probabilities before it starts encoding. The compressor and the decompressor can construct the encoding tree with this probablity information. A dynamic version of Huffman compression can construct an instantaneous Huffman tree while reading and actively compressing. The tree is constantly updated to reflect the changing probabilities of the input data. The coding algorithm proposed by Lempel-Ziv is based on parsing the input data string and considering the resulting phrases as the events to be encoded. This methods does not presuppose any a priori information. The underlying probability structure is learned and exploited in the construction of the code. Önce the input sequence has been obtained, the Lempel-Ziv model structure parses the data string, and a set of phrases is obtained. The phrases ör substrings may be vievred as an extended alphabet on the original symbols. Lempel-Ziv-Welch algorithm has its basis on the Lempel-Ziv algorithm. The procedure equally does not employ a probabilistic source for a data string and effectively counts string symbols within parsed phrases, which are mapped into fixed-length codes. it requires a translation table whose size increases according to the input length. The simplest type of redundancy in the data is long runs of repeating symbols. Run-length encoding algorithm finds runs of these symbols in the data and replaces those runs with a count of original number of symbols and a single symbol. it is based on the repeability of adjacent symbols. Run-length encoding should be applied to the data in which many repeat- viiiing runs occur and it shouldn't be applied to runs of öne ör two symbols. After the researches of lossless compression methods, static Huffman coding is applied to medical images. As the performance of the Huffman coding is measured with the entropy, the entropy of the images, used in this study, are decreased by removing the correlation among the pixels, to increase the compression ratio. Two different decorrelation methods are ap¬ plied to images, first öne is öne of JPEG's (The Joint Photographic Expert Group) predictors, the other is again a predictive decorrelation method, im- proved by Li, Gökmen, Hirschman and Wang. in addition to coding the images, the analysis of the celi images are handled. Öne area of interest to researchers in the biological sciences is the growth and changes of cells. Specially, this can involve such areas as cancer research, proper fooddiets for domestic animals, and efFects of various drugs on metabolisms. The aim of this analysis is to extract the necessary infor- mation, relevant to any illness, from the celi images automatically. Studies done on cells, in otherwords the analysis of the celi images, plays a great role in diagnosis and treatment of the illness. The results of the analysis may contain whatever the doctors want. The automatic analysing processes make the doctors' load to be lighted. in this study, the images are composed of nucleus and nucleolus found in nucleus and the background is styplasma. The doctors have provided that the ratio of the nucleus area, to the sum of the nucleolus area found in the current nucleus carries significant information about the improvement of the illness. Due to the large number of cells in an image and large number of images make this analysing process to be a good candidate for automation. The first step of the analysing process is to segment the nucleus and nu¬ cleolus seperately. The segmentation of these particles clarifies the borders of the particles. it is really hard process for these images, because of the poor contrast between the nucleus and the sytoplasma and the changes in the intensity of the sytoplasm. First of ali, the borders are drawn with a pen and then with mouse. The colour that is used in drawing is a colour that is not found in the images, it is white. But these operationş are far away from automation and take much time. So an other way is used for automatic segmentation, by using the image processing applications. As the images are really hard for the edge detection algorithms, some process is applied to make the images more suitable for these algorithms. To increase the contrast, histogram equalization is applied to the images. a histogram of gray-level content provides a global description of the ap- pearence of an image. Histogram equalization is öne of the ways of achieving enhancement of by modifying the histogram of a given image in a specified manner. The type and degree of enhancement obtained depend on the na- ture of the specified histogram. After the equalization, the nucleus becomes ixmore appearent in the image. To prevent the detecting of variations in the sytoplasma as edge points, the images are thresholded. The characteristic feature of thresholding is a technique widely used in image segmentation. In this applications, thresholding eliminates the undesired borders as much as possible. These operations increase the quality of the images for edge detection algorithms. Three edge detection methods are used, Sobel, Robert and Canny. These methods are based on the gradient operation. In Sobel and Robert methods, an impulse response filter are improved. The convolution of this filters with the images, gives the measurement of the changes in intensity. If there is a rapid change, the magnitude of the gradient is high, or vice-versa. Com paring this magnitude with a threshold, determines the edge points. Canny edge detection method is a computational approach to edge detection. The sucess of this approach depends on the definitions of a comprehensive set of goals to edge points. These goals must be precise enough to delimit the desired behaviour of the detector while making minimal assumptions about the form of the solution. In this method detection and localization criteria are defined and the mathematical representation of these criteria as func tional on the operator of impulse response. There is also an other criteria adding to detection and localization, only one response to a single edge. The numerical optimization of these three criteria is used to derive detec tors for several common image features. The optimal detector has a simple approximate in which edges are marked at maxima in gradient magnitude of a Gaussian-smoothed image. The detecting edges contain undesired edges as well as desired edges. To prevent the accepting of these edges as border points, a cleaning operation is applied. In this operation, the borders that are not composed of closed curves, are eliminated. Before the segmentation, the recent segmentation methods are exam ined. After the segmentation, the segmented particles are labelled. The number of the labels gives the number of the particles is found in the im age. The Blob coloring algorithm is used in labelling. The areas of these particles are calculated by counting the pixels inside the border. As the ratio is the concerning subject the hierarchical structure of the cell images are also defined by detecting which nucleolus is in which nucleus. After the analysis of the cell image, the results are stored in such a way that the analysis will not be repeated in later applications. The nucleus and the nucleolus are represented by squares and circles having the same areas and same hierarchical structures with the original ones. In other words a symbolic representation is obtained. In this study, after achieving the compression of the digital medical im ages in an efficient way, the cell images are analysed automaticaly, as muchas possible, to extract the necessary information. To prevent the repetiton of the analysis in later applications, the symbolic representation is stored. XI
Benzer Tezler
- İnfertil erkeklerde üreme hücrelerinin analizinde digital fotogrametri teknikleri kullanılması üzerine bir çalışma
An investigation on the use of photogrammetric techniques for analysing the reproducing cells of infertility men
LÜTFİYE ALP
Yüksek Lisans
Türkçe
2004
Jeodezi ve FotogrametriSelçuk ÜniversitesiJeodezi ve Fotogrametri Mühendisliği Ana Bilim Dalı
PROF.DR. FERRUH YILDIZ
- Microwave imaging problem with veselago lens structure
Veselago lens yapısı ile mikrodalga görüntülemeproblemi
FURKAN ŞAHİN
Yüksek Lisans
İngilizce
2022
Elektrik ve Elektronik Mühendisliğiİstanbul Teknik ÜniversitesiElektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
PROF. DR. ALİ YAPAR
- Detection of liver fibrosis on slide-level labeled unstained biopsies by quantitative phase imaging and multiple instance learning
Karaciğer fibrozisinin slayt seviyesinde etiketlenmiş boyasız biyopsilerden kantitatif faz görüntüleme ve çoklu örnek öğrenmeyle tespiti
LÜTFİ KADİR ÇELEBİ
Yüksek Lisans
İngilizce
2023
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Teknik ÜniversitesiElektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
DOÇ. DR. İSA YILDIRIM
- Perceptual watersheds for cell segmentation in fluorescence microscopy images
Floresan mikroskop görüntülerinde hücre bölütlemesi için algısal su-seddi algoritması
SALİM ARSLAN
Yüksek Lisans
İngilizce
2012
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİhsan Doğramacı Bilkent ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
YRD. DOÇ. DR. ÇİĞDEM GÜNDÜZ DEMİR
- Deniz-türevi moleküllerin nörovasküler etkilerinin 3b mikro çerçeve ko-kültür modellerinde incelenmesi
Investigation of neurovascular effects of marine-derived molecules in 3d micro frame co-culture model
İREM POLAT
Yüksek Lisans
Türkçe
2019
Biyoteknolojiİzmir Katip Çelebi ÜniversitesiNanobilim ve Nanoteknoloji Ana Bilim Dalı
DR. ÖĞR. ÜYESİ ŞADİYE EMEL SOKULLU