Radiomics is an emerging field in medical image analysis that focuses on extracting a high volume of quantitative features from medical images. It has gained increasing importance due to its ability to non-invasively provide detailed tumor characterization. Given that CT scans are routinely taken throughout the diagnosis and monitoring of a disease, and considering that multiple regions of interest (RoIs) can be segmented from a single image, it is possible to obtain a longitudinal, multi-regional radiomic dataset. In this thesis we develop prognostic models for predicting overall survival and disease-free survival in patients with Colorectal Liver Metastasis (CRLM), utilizing both clinical data and radiomic features. Initially, we introduce a non-parametric survival analysis approach to radiomic data, complemented by a robust feature selection method aimed at identifying all-relevant features, and we compare its performance to traditional Cox models. Then, we quantify the temporal dynamics of tumor response to neoadjuvant chemotherapy by analyzing changes in imaging characteristics from sequential CT scans. Within this framework, we assess the predictive capabilities of our model by incorporating changes in radiomic data over time. Furthermore, we adopt a multi-lesion perspective by quantifying the intra-patient lesion heterogeneity in individuals with CRLM and exploring its prognostic significance for predicting overall survival and disease-free survival. Our objectives are twofold, encompassing both methodological and clinical aims. Methodologically, we seek to determine whether radiomic data from different RoIs and sequential CT scans of the largest lesion can provide valuable prognostic information for patients with CRLM in a non-parametric setting. Additionally, we aim to evaluate whether intra-patient lesion heterogeneity offers complementary insights into the clinical behavior of the disease. Clinically, our goals involve identifying key risk factors associated with imaging variables, identifying patients with similar prognoses and imaging characteristics, and providing a quantitative measure of the treatment effect, moving beyond qualitative evaluations.
La radiomica è un campo emergente nell’analisi delle immagini mediche che si concentra sull’estrazione di un alto volume di caratteristiche quantitative dalle immagini mediche. Ha guadagnato crescente importanza grazie alla sua capacità di fornire in modo non invasivo una dettagliata caratterizzazione dei tumori. Dato che le TAC sono routinariamente effettuate durante la diagnosi e il monitoraggio di una malattia, e considerando che è possibile segmentare molteplici regioni di interesse (RoI) da una singola immagine, è possibile ottenere un dataset radiomico longitudinale e multi-regional. In questa tesi sviluppiamo modelli prognostici per predire la sopravvivenza complessiva e la sopravvivenza libera da malattia nei pazienti con Metastasi Epatiche da Carcinoma Colorettale (CRLM), utilizzando sia dati clinici che caratteristiche radiomiche. Inizialmente, introduciamo un approccio di analisi della sopravvivenza non-parametrico ai dati radiomici, completato da un robusto metodo di selezione delle caratteristiche mirato a identificare le caratteristiche tutte rilevanti, e confrontiamo la sua performance con i modelli Cox tradizionali. Quindi, quantifichiamo la dinamica temporale della risposta del tumore alla chemioterapia neoadiuvante analizzando i cambiamenti nelle caratteristiche di imaging da TAC sequenziali. In questo contesto, valutiamo le capacità predittive del nostro modello incorporando cambiamenti nei dati radiomici nel tempo. Inoltre, adottiamo una prospettiva multi-lesione quantificando l’eterogeneità delle lesioni intra-paziente in individui con CRLM ed esplorando la sua significatività prognostica per predire la sopravvivenza complessiva e la sopravvivenza libera da malattia. I nostri obiettivi sono multipli, abbracciando sia scopi metodologici che clinici. Metodologicamente, cerchiamo di determinare se i dati radiomici da diverse RoI e TAC sequenziali della lesione più grande possono fornire informazioni prognostiche preziose per i pazienti con CRLM in un contesto non-parametrico. Inoltre, vogliamo valutare se l’eterogeneità delle lesioni intra-paziente offre informazioni complementari sul comportamento clinico della malattia.
Non-parametric survival learning for prognostic modeling: integrating temporal imaging data and intra-patient lesions' heterogeneity
Moroni, Sofia
2023/2024
Abstract
Radiomics is an emerging field in medical image analysis that focuses on extracting a high volume of quantitative features from medical images. It has gained increasing importance due to its ability to non-invasively provide detailed tumor characterization. Given that CT scans are routinely taken throughout the diagnosis and monitoring of a disease, and considering that multiple regions of interest (RoIs) can be segmented from a single image, it is possible to obtain a longitudinal, multi-regional radiomic dataset. In this thesis we develop prognostic models for predicting overall survival and disease-free survival in patients with Colorectal Liver Metastasis (CRLM), utilizing both clinical data and radiomic features. Initially, we introduce a non-parametric survival analysis approach to radiomic data, complemented by a robust feature selection method aimed at identifying all-relevant features, and we compare its performance to traditional Cox models. Then, we quantify the temporal dynamics of tumor response to neoadjuvant chemotherapy by analyzing changes in imaging characteristics from sequential CT scans. Within this framework, we assess the predictive capabilities of our model by incorporating changes in radiomic data over time. Furthermore, we adopt a multi-lesion perspective by quantifying the intra-patient lesion heterogeneity in individuals with CRLM and exploring its prognostic significance for predicting overall survival and disease-free survival. Our objectives are twofold, encompassing both methodological and clinical aims. Methodologically, we seek to determine whether radiomic data from different RoIs and sequential CT scans of the largest lesion can provide valuable prognostic information for patients with CRLM in a non-parametric setting. Additionally, we aim to evaluate whether intra-patient lesion heterogeneity offers complementary insights into the clinical behavior of the disease. Clinically, our goals involve identifying key risk factors associated with imaging variables, identifying patients with similar prognoses and imaging characteristics, and providing a quantitative measure of the treatment effect, moving beyond qualitative evaluations.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Moroni_Executive_Summary_02.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Executive Summary
Dimensione
574.01 kB
Formato
Adobe PDF
|
574.01 kB | Adobe PDF | Visualizza/Apri |
2024_04_Moroni_01.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi
Dimensione
4.3 MB
Formato
Adobe PDF
|
4.3 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/219406