engleski

Left Atrial Appendage Segmentation and Analysis in Cardiovascular CT Images

Cardiovascular diseases are the main cause of death both in Europe and globally. Predominantly the elderly population suffers from cardiovascular diseases, especially in the developed countries, while at the same time the world population is getting older. Globally, number of people above 60 is estimated to more than double in thirty years from now. Thus, any improvements in the methods for diagnosis, treatment and prevention of cardiovascular diseases will significantly increase the quality of life and reduce the cost of treatment. Atrial fibrillation is a cardiovascular disease which mostly affects the elderly population and it drastically increases the risk of stroke. The disease is caused when the disorganized electrical signals in the upper heart chambers overwhelm the normal electrical signals propagating through the electrical pathways in the heart. This chaotic electrical activity causes asynchronous contractions and the heart beats outside of the regular sinus rhythm. The asynchronous contractions impede the exchange of blood through the chambers, preventing the complete filling and emptying of the chambers. The blood pools in the chambers, enabling the formation of thrombi. When the thrombi become dislodged and enter the blood circulation (becoming thromboemboli), they can cause stroke. Estimates show that over 90% of strokes caused by cardiovascular diseases are caused by the thromboemboli formed in the left atrial appendage (LAA), small pouch-like structure protruding from the left atrium. A novel percutaneous procedure called the left atrial appendage occlusion has recently been approved for the reduction of the risk of stroke in patients suffering from atrial fibrillation. During the procedure a device called the occluder is placed in the neck of the left atrial appendage, effectively closing it off from the rest of the heart and stopping the blood flow through the LAA. Several occluder device types are available on the market from different manufacturers. Each occluder device type comes in several predefined sizes. Physicians choose a device of the correct size according to each patient’s anatomy. Physicians have to be able to determine the accurate anatomical measurement in order to be able to properly size the device. Advancements in the development of medical imaging modalities, such as computed tomography (CT) or magnetic resonance imaging (MRI), have enabled the acquisition of detailed three-dimensional images of the patients cardiovascular anatomy. Physicians can determine detailed anatomical characteristics of patients’ cardiovascular anatomy from such three-dimensional images. The images can be used for pre-procedural planning of the procedures, reducing the time spent on administering the procedure and reducing the complication rate of the procedure. For example, even though percutaneous LAA occlusion can be administered without the prior CT scan, reports have noted a decrease in total time required for the administration as well as a decrease in the complications during the procedure if the preprocedural CT has been administered. Additionally, the patients prefer the pre-procedural CT scan to the pre-procedural transesophageal echocardiography (TEE), despite the increased irradiation during the imaging. Currently, physicians perform the pre-procedural planning with CT in two main ways: either (1) they measure the patients anatomy directly in 2D slices, or (2) they analyse the 3D model of the LAA. Direct analysis of the 2D slices, even when using the multi-planar reconstrution (MPR), is subjective and error-prone. Certain characteristics of the LAA can be determined differently depending on the plane of the reconstruction, while determining them from the 3D visualization of the LAA is less error-prone and less subjective. Thus, accurate 3D segmentation methods of the LAA are very important for pre-procedural planning. This thesis is focused on the segmentation and the analysis in the cardiovascular CT images in order to reduce the time physicians spend on the pre-procedural planning of the LAA occlusion procedure. The final goal of the thesis is to present the methods which will enable the physicians to – with minimal interaction – determine the feasibility of the procedure for the patient, segment the LAA and determine the location for the placement of the device. The main scientific contributions of this thesis are the three novel methods for the LAA segmentation and analysis, which could improve the preprocedural planning of the occlusion. All presented methods require minimal interaction, as the physician only has to select two parameters in the input CT image: a single pixel (seed point) marking the location of the appendage in one of the slices and a single parameter (threshold) value. Both parameters are intuitive to trained medical users. One of the most important scientific contributions of this work is the method for the centerline detection through the appendage. The detected centerline stretches from the seed point in the appendage to the center of the left atrium. The proposed method detects a centerline in the 3D image by tracking the voxels with the largest radius of the maximum inscribed spheres. The detected centerline is used as an input to the two subsequent methods: the LAA segmentation method and the LAA orifice localization method. However, the reason for the centerline detection is not only to use it as an input in the subsequent methods. The detected centerline allows us to determine the length of the appendage, which is an important parameter for the sizing of the device and an exclusion criteria for a certain types of devices (the ratio of the width and the length of the appendage determines the exclusion criterion for the Watchman and LARIAT devices). Currently the length is determined by a direct measurement in the transesophageal echocardiography (TEE) imaging during the procedure, by a direct measurement in CT slices using the MPR, and finally by specialized software where the physician manually selects the points on the centerline. Our proposed method detects the centerline using only one seed point. Finally, the length of the appendage is calculated from the detected centerline. The second key contribution of this work is the method for the segmentation of the left atrial appendage based on the detected centerline. Left atrial appendage segmentation methods are proving to be increasingly clinically important because they enable the use of different techniques for the pre-procedural planning. One of the most important appendage characteristics is the type of the morphology, which is an exclusion criterion for the procedure in certain types of morphologies. The morphology can be simply determined visually by the physician from the 3D model of the LAA. Determining the morphology type from the 2D slices is error-prone, since the appendage looks differently depending on the angle of the MPR reconstruction. Additionally, accurate segmentation allows for the simple determination of the volume of the appendage, which is another factor in determining the risk of stroke. Finally, proliferation of the 3D printing in the pre-procedural planning, combined with the availability of the accurate LAA segmentation methods, allows the physicians to 3D print the model of the heart and the appendage and correctly determine the correct size of the device prior to the procedure. The proposed segmentation method gradually grows the region marked by the detected centerline and accurately extracts the region containing the LAA and most of the left atrium from an initial mask image created by thresholding the input image. The extraction of the appendage together with the left atrium (LA) area around the appendage allows better understanding of the appendage in the context of the surrounding atrial anatomy (e.g. position and direction of the appendage and proximity to blood vessels). The main advantage of the proposed method is the robustness to the selected threshold value and to the leaks occurring in the mask image. Currently, to the best of our knowledge, very few LAA segmentation methods are available on the market, while the standard region growing methods used in the interactive segmentation software are not robust to leaks after thresholding. The third major scientific contribution of the thesis is the method for the localization of the LAA ostium which uses the detected centerline to determine the plane in 3D space delineating the left atrium from the appendage in the segmentation result from the previous method. The shape of the LAA ostium, determined in the segmented image by intersection with the delineation plane, is an important factor in choosing the type of the device used for the occlusion. Certain types of ostia also indicate a greater risk of peri-device leakage of the blood. Currently, the ostium shape is determined visually in the 2D slices using the double oblique view – MPR centered in the neck of the appendage. By using the proposed method, the physician does not have to modify the MPR planes manually, as the ostium plane is determined by the intersection of the segmented LAA and the determined delineation plane. The ostium shape directly indicates the sizing of the device to be used for the procedure. All three proposed methods are validated against the ground truth segmentations manually created by two medical experts (a radiologist and a cardiosurgeon). The methods achieve large overlap coefficients against the ground truth segmentations. Finally, we have developed an application which enables the physician to visualize the LAA from the input image and easily calculate the required parameters for the procedure. Our work in this area resulted in two published papers in journals in the Science Citation Index and appeared in proceedings of four international conferences.

left atrial appendage ; cardiovascular segmentation and analysis ; cardiovascular CT images

nije evidentirano