LDL cholesterol estimation by computational methods (Master thesis)
Συνεβριώτη, Κωνσταντίνα - Γεωργία/ Synevrioti, Konstantina - Georgia
Introduction: Accurate calculation of LDL cholesterol is extremely important, due to its connection with the occurrence of cardiovascular diseases. Apart from its direct measurement, it is also often calculated by the Friedewald equation, whose accuracy is limited in some cases. Many computational methods have been proposed, the most recent being machine learning. Methods: Machine learning models were trained on patients' datasets, in order to predict LDL cholesterol, using the values of total cholesterol, HDL cholesterol and triglycerides. Those datasets were from two different time periods. Methods used for that were multiple linear regression, support vector regression, xgBoost and deep neural networks. Results: The performance of these models was assessed by the root mean deviation (RMSE). Machine learning methods, and deep neural networks in particular, appeared to outperform Friedewald equation. Efficiency of these models depends on the size and the time interval of the data sets used. Conclusions: Direct measurement of LDL cholesterol seems the best option. But, when it is calculated indirectly, machine learning methods, and specifically deep neural networks, are a reliable solution.
|Institution and School/Department of submitter:||Δημοκρίτειο Πανεπιστήμιο Θράκης. Σχολή Επιστημών Υγείας. Τμήμα Ιατρικής|
|Subject classification:||Low density lipoproteins|
|Keywords:||Low density lipoproteins,Cholesterol,Machine learning,Λιποπρωτεΐνες χαμηλής πυκνότητας,Χοληστερόλη,Μηχανική μάθηση|
|Appears in Collections:||Δ.Π.Μ.Σ. ΒΙΟ-ΙΑΤΡΙΚΕΣ ΚΑΙ ΜΟΡΙΑΚΕΣ ΕΠΙΣΤΗΜΕΣ ΣΤΗ ΔΙΑΓΝΩΣΗ ΚΑΙ ΘΕΡΑΠΕΙΑ ΑΣΘΕΝΕΙΩΝ|
Files in This Item:
|SynevriotiK_2022.pdf||Μεταπτυχιακή εργασία||1.37 MB||Adobe PDF||View/Open|
Please use this identifier to cite or link to this item:This item is a favorite for 0 people.
This item is licensed under a Creative Commons License