Development of X-ray Image-based Navigation Techniques for Percutaneous Coronary Intervention

Abstract

Cardiovascular disease is the world’s leading cause of death announced by the World Health Organization (WHO), the second in Korea. For the treatment of angina pectoris and myocardial infarction, which are representative cardiovascular diseases, more than 50,000 cases of percutaneous coronary intervention (PCI) are performed annually in Korea. Moreover, the number of procedures is gradually increasing with the prevalence of a westernized diet and aging population. Meanwhile, there are no effective preventive measures or treatments for side effects caused by X-rays and contrast agents used in percutaneous coronary intervention. In particular, contrast-induced nephropathy is a common side effect that occurs in 10–30 % of cases after intervention, thereby increasing medical costs, morbidity, and fatality.

In the beginning of the procedure, cardiologists capture angiographic images by using a C-arm X-ray device and contrast agent that show the shape of coronary vessels. Subsequently, they perform the procedure while viewing a live (10–20 Hz) fluoroscopy at a relatively low dose. At this time, a surgical tool called a guidewire is first inserted into the lesion to establish a path for the catheter to enter. In this process, cardiologists manipulate the guidewire by comparing the angiographic images taken in advance with the live fluoroscopic image. As they look at different screens alternately, human errors such as incorrect insertion may occur owing to visual discrepancy, and the procedure time can be long. In addition, a contrast agent is used, which puts a strain on the patient’s kidney, to check the relative position of the coronary vessels and the guidewire. To solve these problems, we herein aim to register angiographic images onto a live fluoroscopic image. To match images captured at different photographing times, an electrocardiogram-based temporal registration method was designed and spatial registration through image processing on the guidewire and coronary vessels was developed. To check the feasibility of our developed methodology, a user study was performed using a phantom vessel model; it helps shorten the procedure time and reduce human errors.

On the contrary, in existing image registration methodology, an electrocardiogram is widely used; electrocardiograms have a problem of causing a malfunction due to a large change in the heartbeat cycle or the presence of arrhythmia or atrial fibrillation. In this thesis, a preliminary study was conducted by devising a frame-matching method based on deep learning to develop an image registration methodology operating independent of the electrocardiogram signals. For matching the frame between angiographic images and the real-time X-ray image, a method to obtain similar frames from angiographic images by using a Siamese neural network that can measure similarity between two inputs is proposed. The basic validity of our proposed method was confirmed through the results of relatively robust operation even with a relatively small amount of label information.

In recent years, robot procedures have been in the spotlight, and procedure robots for percutaneous coronary intervention have been commercialized and used. However, in robotic surgery, cardiologists tend to loss hand–eye coordination and direct sight to the organs. In this study, assuming a future-oriented procedure wherein robotic procedures are emerging, a haptic navigation method is proposed that uses only fluoroscopic and angiographic images without requiring expensive sensorized catheters and guidewires. In our proposed method, once virtual fixtures are generated on the angiographic images in a preprocessing step, the position and tangent vector of the guidewire tip in the live fluoroscopic image can be calculated with respect to the anatomy of the coronary arteries. Finally, haptic feedbacks are generated and transferred to the master robot to help cardiologists manipulate the slave robot directly manipulating the guidewire. A feasibility test with a phantom coronary vessel model and an in-house master–slave robot showed a decrease in task completion time in the error count of wrong insertion after adding haptic navigation. The results showed that haptic navigation was effective for guidewire manipulation.