Bt Görüntüleri Üzerinden Kalp Odacıklarının Bölütlenmesine Yönelik Alternatif Bir Yaklaşım

Abstract

Bu tez kapsamında, Bilgisayarlı Tomografi (BT) görüntüleri üzerinden kalp odacıklarının bölütlenmesine yönelik alternatif bir yaklaşım üzerine çalışılmıştır. Kalp odacıklarının bölütlenmesi, 3 boyutlu kalp modelinin oluşturulmasının temelinde yer almakla beraber, kalbin medikal görüntülenmesinde de büyük önem taşımaktadır. Çalışmada bölütleme işleminin Hounsfield Ünit (HÜ)’ten bağımsız olarak yapılması amaçlanmıştır. Bu amaç doğrultusunda BT görüntüleri üzerinde kümeleme yöntemi ile bölütleme yapılmış ve HÜ’e göre 3 boyutlu bölge genişletme yöntemi ile bölütlenmiş aynı verilerle karşılaştırma yapılmıştır. Üzerinde çalışılan kümeleme yönteminin pencere genişliği (window width ) 600, pencere seviyesi (window level) 80 olan BT görüntülerinde çalışıyor olması sayesinde, pencere genişliği 600, pencere seviyesi 80 olarak kaydedilmiş 8 bitlik BT verileri üzerinde de uygulanabilirliği veri yükü açısından da avantaj sağlamaktadır. Tez çalışması kapsamında iki farklı veri grubu üzerinde çalışılmıştır. Birinci çalışma grubu kalbin hem sağ hem sol odacıklarında kontrast tutucu madde bulunduran, yaş ortalaması 49 olan 4’ü erkek 2’si kadın 6 vakadan oluşmaktadır. İkinci çalışma grubu kalbin sadece sol odacığında kontrast madde bulunduran, yaş ortalaması 62 olan 9’u erkek 11’i kadın 20 vakadan oluşmaktadır. Birinci çalışma grubunun BT görüntüleri üzerinden kalbin hem sağ hem sol odacıkları bölütlenmiştir. İkinci çalışma grubunda ise kalbin sadece sol odacıkları bölütlenmiştir. Hem kümeleme yöntemi hem de 3 boyutlu bölge genişletme yöntemi ile de sağ odacıkların bölütlenmesi gerçekleştirilmiştir. Birinci çalışma grubunda, 3 boyutlu bölge genişletme yöntemiyle bölütlenmiş sağ odacıkların verilerin doğruluk oranı, çok az bir farkla kümeleme yöntemiyle bölütlenmiş sağ odacıkların doğruluk oranına göre daha yüksek çıkarken aynı çalışma grubunda sol odacıkların bölütlenmesinde ise tam tersi kümeleme yöntemi ile bölütlenmiş sol odacıkların doğruluk oranları, 3 boyutlu bölge genişletme yöntemi ile bölütlenmiş sol odacıkların doğruluk oranına göre daha yüksek çıkmıştır. 20 vakadan oluşan ikinci çalışma grubunda ise yine çok az bir farkla kümeleme yöntemine göre bölütlenmiş sol odacıkların doğruluk oranı, 3 boyutlu bölge genişletme yöntemine göre daha yüksek çıkmıştır. Bütün vakalarda genel olarak yöntemlerin doğruluk oranları karşılaştırıldığında ise kümeleme yönteminin doğruluk oranı,3 boyutlu bölge genişletme yöntemine göre daha yüksektir.

In cardiovascular medicine, high-resolution, 3 dimensional monitoring of the patient’s cardiac anatomy is an emerging tool for diagnosis and surgical planning. State-of-the-art cardiac computed tomography scan technology produces a corpus of still 3 dimensional volumetric renditions from the traditional dynamic 2 dimensional images, allowing physicians to envisage the 3 dimensional anatomy of the heart. Cardiac Computed Tomography (CT) is an important diagnosis imaging modality for cardiovascular disease and a method where, detailed anatomic information about the cardiac chambers, large vessels, or coronary arteries can be obtained. Segmentation of cardiac chambers is essential for quantitative analysis, and miscellaneous methods were proposed in the literature. In this study, a new patient-specific segmentation method of the left-right ventricle and atrium was proposed. The proposed algorithm is fast and based on clustering which was developed. In addition to this, developed segmentation algorithm works without using Hounsfield Units (HU). This new approach segmentation method works at 600 window widths and 80 window levels. The new approach segmentation method first finds the minimum and maximum intensity values of data. The distance between minimum and maximum intensity value is divided by the number of the class. The number of class value was determined as 5 classes after the all analysis. The new approach clustering segmentation method was developed for 2 cases and it was applied on standard cardiac computed tomography scans with a contrast agent for a total of 26 cases. Since the left ventricle- atrium and right ventricle- atrium were covered with the contrast agent in 6 cases, the left ventricle- atrium and right ventricle- atrium were segmented. However in the other 20 cases, only the left atrium and ventricle were covered with the contrast agent. Therefore, only the left atrium and left ventricle were segmented. The new approach segmentation method’s results were compared with 3 dimension region growing method’s results. 3 dimensional region growing method is a simple region based segmentation method. In 3 dimensional region growing methods, the user manually chooses seed point from aortic chamber firs and the threshold value is determined as +250 according to Hounsfield Units. The process is iterated on neighbor pixels. If the intensity value of neighbor pixel is above the threshold value, it is including to region. Otherwise the process stops. The image segmentation is a problem that is still one of the main discussion subjects in the field of digital image processing and computer vision. In general, segmentation is a classification of tissues and used for many different tasks such as texture analysis, decision making, data mining and image segmentation. The success of the original segmentation is based to simplify the image and create homogeneous segments. Different classification and segmentation methods were developed in recent years. The threshold, clustering and classification, edge-based segmentation, semantic classification, spectral-segmentation-based classification, region growing, split-and-merge, and mathematical morphology methods are referred to as pixel-based segmentation. In our study, 26 cases were investigated. CT images were taken by an Aquilion 64 Slice at Toshiba Medical Systems, Tokyo, Japan, using standard procedures. 16 bit, 512x512 DICOM (Digital Imaging and Communications in Medicine) retrospective data were used to obtain segmentation results. For the segmentation of the left-right chambers, 6 CT image sets were used where the left-right chambers were interacted with a contrast agent. Segmentation of the left atrium and ventricle for 20 cases were processed, since in these cases contrast agent interaction was only present in the left chambers. .NET platform was used to code the segmentation method and 3 dimensional region growing algorithm. 130 slices were selected randomly from 26 cases and they were used for calculation of true positive (TP), true negative (TN), false positive (FP), false negative (FN), sensitivity (SEN), specificity (SPE), precision (PRE), negative predictive value (NPV) and accuracy (ACC) for two methods, which that are the new approach segmentation method and 3 dimensional region growing. Heart chambers were marked manually in 130 slices by expert radiologist as a blind reader for evaluation as well as analysis of obtained results. The average results for sensitivity, specificity, precision, negative predictive value and accuracy were calculated for both methods. In the presented study we obtained accurate results for heart chambers with the new approach segmentation method. The approach uses standard Computed Tomography Protocols. The segmentation of heart chambers is essential for diagnosis, clinical research and gives to physician necessary information for quantitative functional analysis of the whole heart. The presented study showed that the new approach segmentation method to segment heart chambers was very efficient, and created patient-specific heart segments in CT images. When segmentation results were compared with the findings of physicians, the new approach segmentation method’s average accuracy was 99.26 % for left atrium-ventricle, 98.82 % for right atrium-ventricle segmentation for the first group (6 cases) segmentation and 99.49 % for left atrium-ventricle for the second group (20 cases) segmentation, respectively. 3 dimensional region growing method’s average accuracy was 99.19 % for left atrium-ventricle, 98.87 % for right atrium-ventricle segmentation for the first group (6 cases) segmentation and 99.42 % for left atrium-ventricle for the second group (20 cases) segmentation. General accuracy have been calculated 99.20 % for the new approach segmentation method and 99.16 % for 3 dimensional region growing segmentation method. When you look overall accuracy analysis, though there was not a meaningful difference between these two methods, the accuracy rates of the new approach segmentation method obtained independent from Hounsfield Units were higher than 3 dimension region growing method. The new approach segmentation method works without Hounsfield Units, unlike conventional segmentation methods. In addition, the new approach segmentation method works on 600 window widths and 80 window levels of DICOM image window settings. Hence, this new approach segmeantation method can be used on 8-bits data which saved as .jpg, .png., .bmp, etc. from DICOM image 600 window widths and 80 window levels. This new approach segmentation method creates an alternative way to saving high capacity medical data. When segmentation results were compared with the findings of physicians, the new approach segmentation method’s average accuracy was 99.21 % for left atrium-ventricle, 98.05 % for right atrium-ventricle segmentation for the first group (8 bits, 3 cases) segmentation and 99.19 % for left atrium-ventricle for the second group (8 bits, 3 cases) segmentation, respectively.