Tomografide temel beyin tanılarının otomatik tespiti
Yükleniyor...
Dosyalar
Tarih
2017
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Bursa Teknik Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Bilgisayarlı tomografi (BT) sağlık kuruluşlarının görüntüleme bölümlerinde manyetik rezonans (MR) ile beraber en sık kullanılan kesitsel tanı ünitesidir. Tomografi ve MR cihazlarının şüphe edilen hastalık türüne göre tanı koymada birbirlerine üstünlükleri mevcuttur. Bu nedenle hekimler tanılama için cihaz seçimlerini ön bulgulara göre yapmaktadır. Tomografi cihazının özellikle kanama durumlarında ve kemik kırıklarında üstünlüğü mevcuttur. Buna bağlı olarak acil hizmetlerde tomografi ünitesi MR ünitesine göre daha yoğun kullanılmaktadır. Tanı süreci; hastanın ilgili üniteye yönlendirilmesi ile başlar. Sonrasında ilgili ünitede radyoloji teknikerleri tarafından çekimin gerçekleştirilir. En sonunda oluşan görüntüler radyoloji uzmanı tarafından yorumlanır ve yorumların diktasyonu sağlanır. Diktasyon sonucu oluşan rapor, tanılama sürecinin neticesidir. Her bir hasta için süreç bu şekilde baştan sona ilerleyerek tamamlanır. Tanılama sürecinin hızı ve doğruluğu hayati önem arz etmektedir. Bu tezde, bilgisayarlı tomografi cihazında elde edilen beyin görüntülerinin otomatik olarak işlenmesi ve acilde en sık görülen temel tanılar olan kanama ve enfarkt tanılarının otomatik olarak tespitini sağlayacak bir yazılım oluşturulması üzerinde çalışılmıştır. Çalışmada, mevcut görüntü işleme ve makine öğrenmesi algoritmaları kullanılarak tomografi görüntülerinden tanı yapılabilmesi incelenmiştir. Temel olarak, ilk aşamada tomografi görüntülerinin, sayısal görüntü işleme teknikleri ile öncelikle iyileştirilmesi ele alınmıştır. İyileştiren görüntülerden daha sonra yine farklı görüntü işleme yöntemleri ile tanı yapılabilmesi amacı ile öznitelik çıkarma yöntemleri incelenmiştir ve otomatik tanı yapılabilmesine en uygun öznitelik parametrelerinin seçilmesi gerçekleştirilmiştir. Son aşamada ise elde edilen öznitelikler ile sınıflandırma işlemi uygulanarak bilgisayarın otomatik olarak karar vermesi sağlanmıştır. Bu noktada literatürde en sık kullanılan sınıflandırıcı algoritma olan destek vektör makineleri (SVM) sınıflandırıcısı kullanılmıştır. Deneyler sırasında tamamen normal beyin bilgisayarlı tomografi görüntülerinin yanı sıra kanama ve enfarkt tanılarına sahip gerçek görüntüler veritabanı olarak kullanılmıştır. Öznitelik belirleme için lokal ikili örüntü (LBP) yöntemi tercih edilmiştir. Görüntü işlemede kullanılacak dönüşümler ve öznitelikler için optimal parametreler belirlenmiştir. Tüm test ve işlemler MATLAB programı vasıtasıyla yapılmıştır. Testler kapsamında görüntü iyileştirme yöntemlerinin uygulanmadığı ve her birinin tekil ve farklı kombinasyonlarının uygulandığı tüm durumlar ele alınmıştır. Gerçekleştirilen testlerde kanama vak'aları için en yüksek %67 doğru tespit oranına, enfarkt vak'aları için ise en yüksek %81 doğru tespit oranına erişilmiştir.
Computed tomography (CT) is the most commonly used sequential diagnostic unit with magnetic resonance imaging (MR) in imaging sections of healthcare facilities. Tomography and MR devices have superiority in diagnosing each other according to the suspected disease type. For this reason, clinicians make device selection for diagnosis according to preliminary findings. The tomography device is particularly advantageous in cases of bleeding and bone fractures. Therefore, in emergency services, the tomography unit is used more intensely than the MR unit. The diagnostic process begins with the patient being directed to the relevant unit. Subsequently, the radiology technicians do the scan in the relevant unit. Finally, the images are interpreted by the radiologist and the dictation of the interpretation is provided. The result of the dictation is the result of the diagnostic process. For each patient, the process is thus completed from the beginning to the end. The speed and accuracy of the diagnostic process is vital for patients. In this thesis, automatic processing of brain images obtained from a computed tomography device and creation of a software that will automatically detect the most common basic diagnoses, hemorrhage and infarct, are studied. In the study, the ability to make an automatic diagnosis from tomography images using existing image processing and machine learning algorithms has been investigated. Basically, first of all, the improvement of the tomography images with digital image processing techniques is discussed. Then, the methods of extracting features with the aim of diagnosing with different image processing methods have been examined from the improving images and the most suitable parameters have been selected for automatic diagnosis. At the last stage, classification is performed with the obtained attributes and the computer is automatically decided. At this point, the support vector machines (SVM) classifier, which is the most frequently used classification algorithm in the literature, is used. During the experiments, true brain computed tomography images as well as actual images with hemorrhage and infarcts were used as a database. The local binary örüntü (LBP) method is preferred for feature extraction. Optimal parameters are specified for the transformations and features to be used in image processing. All tests and operations were done by the MATLAB program. Histogram equalization, contrast adjustment, median filtering, average filtering, gamma correction and power-law transform are used as image enhancement methods in the tests. For gamma correction and power-law transform, the application results for different gamma and n values were examined, optimal gamma and n values were determined and these values were used in the tests. Optimal neighborhood method and parameter for LBP are determined by accuracy of results with applying different methods and parameters for LBP. Tests are performed by turns. For each turn, 25 images were selected randomly for training and the remainder were divided into test. After 2000 turns, average of detection accuracy of turns is taken as a result. The tests covered all cases in which image enhancement methods were not applied and the singular and different combinations of each were applied. In the tests performed, the highest detection rate was reached at 67% for hemorrhage cases and at 81% for infarct cases.
Computed tomography (CT) is the most commonly used sequential diagnostic unit with magnetic resonance imaging (MR) in imaging sections of healthcare facilities. Tomography and MR devices have superiority in diagnosing each other according to the suspected disease type. For this reason, clinicians make device selection for diagnosis according to preliminary findings. The tomography device is particularly advantageous in cases of bleeding and bone fractures. Therefore, in emergency services, the tomography unit is used more intensely than the MR unit. The diagnostic process begins with the patient being directed to the relevant unit. Subsequently, the radiology technicians do the scan in the relevant unit. Finally, the images are interpreted by the radiologist and the dictation of the interpretation is provided. The result of the dictation is the result of the diagnostic process. For each patient, the process is thus completed from the beginning to the end. The speed and accuracy of the diagnostic process is vital for patients. In this thesis, automatic processing of brain images obtained from a computed tomography device and creation of a software that will automatically detect the most common basic diagnoses, hemorrhage and infarct, are studied. In the study, the ability to make an automatic diagnosis from tomography images using existing image processing and machine learning algorithms has been investigated. Basically, first of all, the improvement of the tomography images with digital image processing techniques is discussed. Then, the methods of extracting features with the aim of diagnosing with different image processing methods have been examined from the improving images and the most suitable parameters have been selected for automatic diagnosis. At the last stage, classification is performed with the obtained attributes and the computer is automatically decided. At this point, the support vector machines (SVM) classifier, which is the most frequently used classification algorithm in the literature, is used. During the experiments, true brain computed tomography images as well as actual images with hemorrhage and infarcts were used as a database. The local binary örüntü (LBP) method is preferred for feature extraction. Optimal parameters are specified for the transformations and features to be used in image processing. All tests and operations were done by the MATLAB program. Histogram equalization, contrast adjustment, median filtering, average filtering, gamma correction and power-law transform are used as image enhancement methods in the tests. For gamma correction and power-law transform, the application results for different gamma and n values were examined, optimal gamma and n values were determined and these values were used in the tests. Optimal neighborhood method and parameter for LBP are determined by accuracy of results with applying different methods and parameters for LBP. Tests are performed by turns. For each turn, 25 images were selected randomly for training and the remainder were divided into test. After 2000 turns, average of detection accuracy of turns is taken as a result. The tests covered all cases in which image enhancement methods were not applied and the singular and different combinations of each were applied. In the tests performed, the highest detection rate was reached at 67% for hemorrhage cases and at 81% for infarct cases.
Açıklama
Anahtar Kelimeler
Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering, Radyoloji ve Nükleer Tıp, Radiology and Nuclear Medicine
