Exploring new PET/CT capabilities and machine learning for improving the diagnosis of infective endocarditis

This thesis aims to enhance the diagnosis of infective endocarditis: a challenging disease involving the heart's inner lining or implanted materials like artificial valves or cardiac implanted electronic devices (CIEDs). The research primarily focuses on the role of [18F]FDG PET/CT, exploring its current clinical use, particularly after the 2015 ESC guidelines integrated it as a major criterion for suspected prosthetic valve endocarditis.
An overview of infective endocarditis, its history, diagnosis, and treatment challenges introduces the thesis. Subsequent chapters examine the indications for of [18F]FDG PET/CT for suspected infective endocarditis and the need for standardisation of this technique; the benefits of motion corrected FDG PET/CT for improved image quality, and the significance of mediastinal lymph node activity - which was unfortunately unreliable as an indicator for endocarditis.
The thesis also includes a systematic review and meta-analysis on [18F]FDG PET/CT for infections involving Left Ventricular Assist Devices (LVADs), highlighting its high sensitivity and specificity for these infection. They include a proposal for standardizing diagnostic criteria for LVAD infections.
The effectiveness of [18F]FDG PET/CT's was also evaluated in a two-centre cohort study, emphasizing the complementary role of semi-quantitative analysis next to visual analysis in accurately identifying LVAD infections.
Finally, it presents a proof-of-concept on using machine learning to enhance the modified Duke criteria's predictive power for endocarditis. These models showed promising results, though they would require careful further validation before clinical implementation could be considered.