MEDICAL INFORMATICS IV
GENERATIVE ADVERSARIAL NETWORK AND CONVOLUTIONAL NEURAL NETWORKS IN PREDICTION OF CAROTID ARTERY DISEASE PROGNOSIS
Yayıncı:
İstanbul Üniversitesi Yayınları
It is of great importance to ensure patient management by predicting the disease prognosis in terms of early diagnosis of chronic, progressive carotid artery diseases, implementation of treatment options, and prevention or reduction of mortality, morbidity, and complication risks in long-term follow-up. Atherosclerotic carotid artery diseases are the most important cause of stroke, and the risk of stroke increases in proportion to the degree of stenosis. The prevalence of carotid stenosis in the general population is 3%, it increases with advanced age and is more common in men than women. The gold standard treatment for symptomatic disease is carotid endarterectomy. The main goal of surgery is to minimize the risk of recurrent stroke from an unstable plaque. The role of surgery in asymptomatic disease is controversial. Therefore, it is of great importance to predict what kind of disease structure may lead to carotid artery pathologies in the future. Carotid Doppler ultrasound (DUS) and computed tomographic (CT) angiography are imaging techniques that are used to diagnose carotid artery disease. These techniques are particularly helpful in estimating the degree of stenosis. It is crucial to study the different types of vascular structure that will emerge through the modeling of existing images using a machine-learning and image-processing algorithm that supports the identification of early diagnosis from the analysis of carotid artery imaging data. It is crucial to report the results in retrospect and to investigate the aging of the vascular system in the future based on these findings and to create a prototype for a decision support application that could be used to predict the onset of prospective early detection. Convolutional neural networks (CNN) are used for classification and segmentation of images. Creating generative adversarial network (GAN) models can be shown as generating new data with similar properties from a datum. Apart from classification, the GAN method will also be used to generate synthetic data. Thus, it is aimed to produce the development of atherosclerosis synthetically as a time series, to develop the diagnostic support system from the synthetically created image, to train the GAN network, and to train the CNN model. Prognosis prediction of carotid artery diseases with GAN method, long-term follow-up and remote patient management methods to determine risk factors for the disease, to control and reduce them, to carry out regular treatment, to predict the need for urgent or urgent vascular intervention, to prevent adverse events. These methods contribute to reducing the rate and duration of hospitalization, to alleviate the great burden on the health economy due to complications in chronic diseases, and to increase the quality of life of patients. In conclusion, the GAN method will contribute to the risk scoring of cerebrovascular diseases, the prediction of the success and effects of the treatment modalities (preventive, medical, interventional, surgical) with the diagnostic classification approach, the creation of a long-term uncomplicated survival and also the application of personalized medicine.
