National and Kapodistrian University of Athens
Σχεδιασμός και ανάπτυξη συστήματος υπολογιστικής ευφυΐας για την εκτίμηση κινδύνου σε παθήσεις που σχετίζονται με το καρδιαγγειακό σύστημα
Abstract
dc:descriptionAccording to the World Health Organization, cardiovascular diseases, including atherosclerotic coronary heart disease, are the leading cause of death worldwide. The purpose of this doctoral thesis was the digital data analysis, derived mainly from medical images, as well as from biological samples of patients with known or suspected cardiovascular diseases, in order to identify important biomarkers, which can predict outcomes of these entities. Machine learning, through classification and regression algorithms, plays a key role in solving the research queries of this thesis. To perform reading and custom visualization of positron emission tomography (PET)/computed tomography (CT) images an in-house stand-alone software was developed. This facilitated a detailed analysis of the imaging data used to study the research questions posed in this thesis. Initially, the change in radiopharmaceutical uptake in the arterial network over time was studied and predictive models for estimating standardized uptake value (SUV) measurements at different time points (T1, T2) were implemented. Biomedical data were collected from 45 patients who were free of malignant or inflammatory disease, when they underwent 18-fluorodeoxyglucose (18F-FDG) PET/CT imaging. Linear and non-linear regression models of the SUV difference (ΔSUV) and the time difference ΔΤ=Τ2-Τ1 were developed based on the derivation group (30 patients) and tested in an independent validation group (15 patients). The models demonstrated very good reproducibility with minor non-statistically significant differences, likely of limited clinical importance, between measured and the predicted SUVs. Another scientific query that was approached in the present thesis was to identify biomarkers predicting high risk atherosclerotic disease. In two different sets of patients, who had a high-grade carotid stenosis and undergone carotid endarterectomy, texture analysis was studied for the first time in 18F-FDG PET/CT carotid imaging (1st set) and microarrays analysis of gene expression data containing core and shoulders of the plaque and distant macroscopically intact tissue (2nd set), taken from the Gene Expression Omnibus (GEO) online database. Sixteen features based on intensity histogram and 51 features based on various analytical matrices that already encode specific spatial relationships between the pixels and characterize the spatial heterogeneity were extracted as imaging biomarkers. Differentially Expressed Genes (DEGs) were sorted based on the absolute value of the log Fold Change (logFC) and used individually as inputs to classifiers for the purpose of identifying potential biomarkers. After analyzing the data, through machine learning algorithms, it was shown that texture analysis of PET carotid images provides valuable complementary information regarding biological plaque characteristics compared to conventional imaging markers identifying tracer concentration within the area of a carotid lesion that may not be captured by the human eye. At the same time, crucial genes have been identified that are involved in the formation of atheroma plaque and may act as therapy targets and prognosis indicators. Then, the ability of machine learning algorithms for predicting stress induced myocardial perfusion changes by PET and Single Photon Emission Tomography (SPECT) imaging was investigated, by combining parameters extracted from Computed Tomography Coronary Angiography (CTCA), such as those based on computational fluid dynamics (CFD) studies, quantitative atherosclerotic plaque burden, plaque morphology and anatomical characteristics, as well as texture features. The results showed that machine learning seems to be useful for predicting myocardial perfusion abnormalities downstream a coronary lesion. Among the best feature combinations, the CDF derived endothelial shear stress (ESS) in the proximal segment of a coronary lesion was present in all classification schemes and in every combination of best features. In addition, plaque anatomy and the consequent changes in fluid dynamics, as captured by ESS, appear to be important indices of flow limiting coronary lesions. In addition, ESS results in an incremental value to stenosis severity by improving the prediction of impaired myocardial perfusion assessed by PET-MPI (Myocardial Perfusion Imaging). Texture analysis can be used in conjunction with CTCA imaging and potentially provide valuable complementary information combined with anatomical and morphological characteristics for predicting inducible perfusion abnormalities on SPECT-MPI, which may not be captured by the human eye.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kafouris, Pavlos
- Καφούρης, Παύλος
Subjects
dc:subject × 21- Αθηρωματική στεφανιαία νόσος
- 18F-FDG PET/CT απεικόνιση
- Μηχανική μάθηση
- Ανάλυση υφής εικόνας
- Βιοδείκτες
- Atherosclerotic coronary heart disease
- 18F-FDG PET/CT imaging
- Machine learning
- Image texture analysis
- Biomarkers
- Βιοϊατρική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμες Υγείας
- Ιατρική και Επιστήμες Υγείας
- Ιατρική Πληροφορική
- Medical Engineering
- Engineering and Technology
- Biomedical Engineering
- Health Sciences
- Medical and Health Sciences
- Medical Informatics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/50153