Chronic low back pain is a leading cause of disability worldwide, but most cases lack a pathoanatomical source on conventional imaging. The role of low-grade inflammation in chronic pain is increasingly understood. Hence, functional imaging techniques such as 18F-Fluorodeoxyglucose positron emission tomography (18F-FDG PET) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) may aid in the diagnosis of LBP by quantifying metabolic and microcirculatory aspects of inflammation, respectively. However, the relationship between these modalities in the lumbar spine remains unclear.
This thesis investigates the correlation between quantitative parameters of PET and DCE-MRI in the chronically painful lumbar spine. The secondary objective is to determine whether and how these correlations vary across different anatomical structures.
18F-FDG PET/DCE-MRI images from 23 patients with chronic LBP were processed using an in-house developed software pipeline (SPARCK) with custom additions/modifications. Regions of interest (ROIs) were manually segmented in eight lumbar structures in areas of increased 18F-FDG uptake and matched reference regions. PET parameters were SUVmean and SUVmax. Pharmacokinetic modeling with the Extended Tofts model was applied to obtain median and peak Ktrans, vp and k¬ep from DCE-MRI. Correlations were tested using rank-based linear mixed-effects model, accounting for repeated measures and skewed distributions. Significance levels were corrected for multiple testing.
Across the total of 179 ROIs, weak but significant positive correlations were observed between SUV and Ktrans metrics (ρLME ≈ 0.31–0.37), and between SUV and vp metrics (ρLME = 0.23) except SUVmean/vp,peak. There were no significant correlations between SUV and kep metrics across all ROIs. Structure-specific analyses showed varying correlations between SUV and Ktrans metrics: strong correlations in the intervertebral disc, moderate ρLME in the spinal canal and facet joint, no correlations in the nerve root and interspinous tissue.
Correlations between quantitative 18F-FDG PET and DCE-MRI parameters in chronic LBP are weakly positive overall and vary across anatomical structures. These results suggest that the imaging modalities capture complementary rather than redundant aspects of inflammation and that their diagnostic value may depend on the anatomical structure under investigation. Future studies should validate these findings in larger cohorts and establish baseline parameter values with asymptomatic controls.

Diagnostic errors remain a major challenge in modern medicine, particularly in emergency departments (EDs), where clinicians operate under high pressure and limited time. Bayesian Networks (BNs) offer a transparent, probabilistic approach to reasoning under uncertainty, which may help reduce these errors. This thesis investigates the development and optimization of Promedas (PRObabilistic MEdical Diagnostic Advisory System), a BN-based clinical decision support system (CDSS) designed to generate accurate differential diagnoses for ED patients.

Building on previous iterations, this work aims to determine whether Promedas’ diagnostic accuracy could approach or surpass that of clinicians. Because access to real patient data was unavailable, a large synthetic dataset was generated using Microsoft Copilot, following realistic clinical templates. This dataset was automatically translated from unstructured clinical text into standardised “Promedas language,” consisting of findings, conditions, and diagnoses. The translation process was optimized using prompt-engineering strategies and validated through a custom reviewer environment, achieving an estimated accuracy of 97.5%.

An optimisation cycle was then developed to iteratively refine Promedas’ configuration, and several thought experiments explored its theoretical performance boundaries, including “best guess” and “best case” strategies. While the system demonstrated structural consistency and technical feasibility, the absence of real-world data limited the interpretability of its quantitative results.

The findings suggest that Promedas, under ideal and fully validated circumstances, holds substantial potential transparent, evidence-based diagnostic reasoning and could complement clinical decision-making. However, its true capability can only be determined through further experimentation and re-evaluation on real patient data. Future work should focus on this experimentation and on finding a reliable workflow to assist the iterative refinement cycle.

Background: Histopathological examination in the diagnostic workflow of oropharyngeal squamous cell carcinoma (OPSCC) is essential. We aimed to develop a machine learning pipeline to predict human papillomavirus (HPV) status in OPSCC patients based on clinical variables and multiparametric magnetic resonance imaging (MRI).
Methods: In a dataset of OPSCC patients (n=59), we extracted features from three categories: clinical variables; histogram parameters from diffusion weighted imaging (DWI)-MRI model; radiomics based on T2-weighted and DWI-MRI. We performed ten-times repeated stratified five-fold cross-validation and divided each outer training set (80%) into an inner training set (80% of outer training set) and validation set (20% of outer training set) using five-fold stratified cross-validation. We performed three types of feature selection methods (LASSO, statistical analysis and manual selection), tuned and trained seven classifiers (logistic regression, k-nearest-neighbours, naive bayes, random forest, support vector machine, XGBoost and LightGBM) to find the optimal combination of features, hyperparameters and classifier on each inner training set. We ensembled the inner fold models to fit on the outer training set and tested on the outer test set (20%). We constructed additional models with subsets of the features.
Results: the combined model area under the curve was 0.793±0.136. Models including clinical features outperformed models without clinical features (p<0.001). Features from all feature categories were selected for the combined model.
Conclusion: we were able to predict HPV status in OPSCC patients using multiparametric MRI and clinical variables with reasonable accuracy, though retraining and validating on larger, external datasets is needed before implementation in clinic.