Novel Applications of Machine Learning to Guide Diagnosis of Endocarditis
(2025)Department of Automatic Control
- Abstract
- Infective endocarditis (IE) is a severe disease affecting the inner lining of the heart. Its diverse symptoms complicate diagnosis and the current main diagnostic tool is the Duke criteria which classifies suspected cases as definite, possible or rejected IE. A substantial portion of cases falls into the ambiguous “possible” category and there is a need for a more precise diagnostic system.
This thesis aimed to investigate the usage of machine learning (ML) models, utilizing non-linear relationships, as a tool to determine whether a patient suffers from IE and extract what features are important in decision making. Multiple ML algorithm techniques were trained with a set of variables extracted from medical records. For target variable... (More) - Infective endocarditis (IE) is a severe disease affecting the inner lining of the heart. Its diverse symptoms complicate diagnosis and the current main diagnostic tool is the Duke criteria which classifies suspected cases as definite, possible or rejected IE. A substantial portion of cases falls into the ambiguous “possible” category and there is a need for a more precise diagnostic system.
This thesis aimed to investigate the usage of machine learning (ML) models, utilizing non-linear relationships, as a tool to determine whether a patient suffers from IE and extract what features are important in decision making. Multiple ML algorithm techniques were trained with a set of variables extracted from medical records. For target variable the opinion from a team consisting of medical specialists of multiple disciplines was used.
The results showed that all ML algorithms outperformed logistic regression, which was used as a linear model for comparison. The Duke criteria in turn outperformed the ML models in specificity and sensitivity. The two best performing models were based on tabular prior-data fitted network and random forest. The ML models largely utilized known predictors of IE but also identified new important variables including estimated glomerular flow rate, C-reactive peptide, no known focus and body mass index. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9212671
- author
- Svenningsson, Axel
- supervisor
- organization
- year
- 2025
- type
- M2 - Bachelor Degree
- subject
- report number
- TFRT-7672
- other publication id
- 0280-5316
- language
- English
- id
- 9212671
- date added to LUP
- 2025-09-18 14:19:05
- date last changed
- 2025-09-18 14:19:05
@misc{9212671,
abstract = {{Infective endocarditis (IE) is a severe disease affecting the inner lining of the heart. Its diverse symptoms complicate diagnosis and the current main diagnostic tool is the Duke criteria which classifies suspected cases as definite, possible or rejected IE. A substantial portion of cases falls into the ambiguous “possible” category and there is a need for a more precise diagnostic system.
This thesis aimed to investigate the usage of machine learning (ML) models, utilizing non-linear relationships, as a tool to determine whether a patient suffers from IE and extract what features are important in decision making. Multiple ML algorithm techniques were trained with a set of variables extracted from medical records. For target variable the opinion from a team consisting of medical specialists of multiple disciplines was used.
The results showed that all ML algorithms outperformed logistic regression, which was used as a linear model for comparison. The Duke criteria in turn outperformed the ML models in specificity and sensitivity. The two best performing models were based on tabular prior-data fitted network and random forest. The ML models largely utilized known predictors of IE but also identified new important variables including estimated glomerular flow rate, C-reactive peptide, no known focus and body mass index.}},
author = {{Svenningsson, Axel}},
language = {{eng}},
note = {{Student Paper}},
title = {{Novel Applications of Machine Learning to Guide Diagnosis of Endocarditis}},
year = {{2025}},
}