Pisani, Noemi (2024) High-Dimensional Quantitative Approach to Texture Analysis of MRI in Parkinsonian Disorders. [Tesi di dottorato]

Documento PDF Thesis_NoemiPisani_FINAL.pdf Visibile a [TBR] Amministratori dell'archivio Download (36MB) | Richiedi una copia

Abstract

Radiomics is an emerging and promising field that tries to quantify information within medical images through the extraction and analysis of radiomic features. This quantitative approach has shown significant potential for predictive and prognostic applications, especially in the context of neurodegenerative diseases, where the main challenge is the identification of in vivo biomarkers. Neurodegenerative diseases often present with overlapping clinical features that complicate their recognition and thus accurate diagnosis, and this is where radiomics could play a crucial role. However, radiomics faces inherent challenges, particularly with regard to reproducibility and robustness of extracted features, which must be addressed to ensure clinical applicability and accuracy. This thesis explores the performance of radiomics in distinguishing atypical forms of parkinsonism and their respective phenotypic variations. By combining radiomic analysis with supervised and unsupervised machine learning algorithms, the study evaluates the ability of these approaches to distinguish between Parkinson's disease and multisystem atrophy, as well as to distinguish phenotypes within progressive supranuclear palsy. The findings reveal that radiomics, when combined with machine learning, shows promising results in classification, emphasising the utility of texture-based features over first-order statistical features to achieve accurate differentiation. Furthermore, this thesis investigates the hypothesis that the choice of a method for normalising the intensities of MRI images may have an impact on the repeatability of radiomic features. Through experimentation with three distinct intensity normalisation techniques applied to MRIs acquired under the same conditions (e.g., same scanner and protocol), the study finds notable variability in the repeatability of features. This insight highlights the need for careful methodological standardisation image pre-processing in radiomic studies to ensure reliable outcomes. Ultimately, this research supports the potential of radiomics as a valuable tool for clinicians in supporting the diagnosis of parkinsonian disorders. While conventional MRI does not always reveal abnormalities in specific brain regions and other advanced imaging techniques are not routinely available, radiomic analysis offers a layer of quantifiable information that could complement qualitative clinical assessments.

Downloads

Actions (login required)

Modifica documento