In recent years, the concept of precision policy in health has gained importance, emphasizing the role of individual-level data in refining population-level healthcare strategies. This thesis explored two distinct applications with a common goal: enhancing disease understanding and supporting clinical decision-making through meaningful patient grouping. The first part of this thesis focused on Colorectal Liver Metastasis (CRLM), a common progression in colorectal cancer, where malignant cells spread from the colon or rectum to the liver. This type of cancer is associated with complex tumor traits and diverse survival outcomes, making it challenging to manage. Radiomics, by extracting numerous quantitative features from CT images, provides a detailed, non-invasive means of analyzing tumor properties. A multi-view perspective was obtained by segmenting multiple regions of interest (RoIs) within a single image. We applied traditional unsupervised techniques and an innovative semi-supervised method incorporating survival data to identify patient subgroups sharing not only radiomic similarities but also analogous survival outcomes. These valuable insights enable clinicians to tailor treatment plans, addressing a primary concern in oncology: understanding survival prospects alongside tumor traits. The second part of this study shifted to heart failure, a common cardiovascular disease with growing prevalence due to the aging population. Treatment for this condition is largely centered on drug therapy, though adherence to these therapies is often low among heart failure patients, which results in worsened health outcomes and increased hospitalizations. As a result, this pathology has a significant impact from both an epidemiological and economic standpoint. Using administrative data from the Lombardy region, this part aimed to identify risk factors and stratify patients into interpretable, survival-guided risk levels. Here, the views were artificially designed to capture diverse facets of patient management, providing a comprehensive overview of the patient’s clinical profile. This approach supports the development of precision policies, offering clinicians a structured framework to assess and manage patient risk more effectively at the population level.
Negli ultimi anni, il concetto di politiche di precisione in ambito sanitario ha acquisito importanza, sottolineando il ruolo dei dati a livello individuale nel migliorare le strategie sanitarie a livello di popolazione. Questa tesi esplora due applicazioni con l'obiettivo comune di migliorare la comprensione delle malattie e supportare le decisioni cliniche tramite una suddivisione mirata dei pazienti. La prima parte si concentra sulle metastasi epatiche da tumore colorettale (CRLM), una progressione comune del cancro del colon-retto, in cui le cellule maligne si diffondono al fegato. La radiomica, estraendo numerose caratteristiche quantitative dalle TAC, offre un mezzo non invasivo per analizzare le proprietà tumorali. Una prospettiva multi-view si ottiene segmentando più regioni d'interesse (RoI) all'interno di una singola immagine. Sono state applicate tecniche di clustering tradizionali non supervisionate e un innovativo metodo semi-supervisionato che ha integrato dati di sopravvivenza, identificando gruppi di pazienti con somiglianze radiomiche ed esiti di sopravvivenza analoghi. Questi risultati permettono ai clinici di personalizzare i piani terapeutici, rispondendo alla necessità di comprendere le prospettive di sopravvivenza in relazione ai tratti tumorali. La seconda parte si focalizza sull'insufficienza cardiaca, una patologia cardiovascolare comune, la cui prevalenza aumenta con l'invecchiamento della popolazione. Il trattamento si basa sulla terapia farmacologica, ma la bassa aderenza tra i pazienti spesso peggiora gli esiti e aumenta le ospedalizzazioni. Questa patologia ha un impatto rilevante sia epidemiologico che economico. Utilizzando dati amministrativi della regione Lombardia, sono stati identificati i fattori di rischio e classificati i pazienti in livelli di rischio interpretabili, guidati dalla sopravvivenza. Qui le view sono state create artificialmente per rappresentare diversi aspetti della gestione del paziente, fornendo un quadro clinico completo del paziente. Questo approccio supporta lo sviluppo di politiche di precisione, offrendo ai clinici un quadro strutturato per valutare e gestire il rischio dei pazienti in modo più efficace a livello di popolazione.
Multi-view clustering methods in clinical research: a comparative assessment of unsupervised and semi-supervised approaches
Iapaolo, Valeria
2023/2024
Abstract
In recent years, the concept of precision policy in health has gained importance, emphasizing the role of individual-level data in refining population-level healthcare strategies. This thesis explored two distinct applications with a common goal: enhancing disease understanding and supporting clinical decision-making through meaningful patient grouping. The first part of this thesis focused on Colorectal Liver Metastasis (CRLM), a common progression in colorectal cancer, where malignant cells spread from the colon or rectum to the liver. This type of cancer is associated with complex tumor traits and diverse survival outcomes, making it challenging to manage. Radiomics, by extracting numerous quantitative features from CT images, provides a detailed, non-invasive means of analyzing tumor properties. A multi-view perspective was obtained by segmenting multiple regions of interest (RoIs) within a single image. We applied traditional unsupervised techniques and an innovative semi-supervised method incorporating survival data to identify patient subgroups sharing not only radiomic similarities but also analogous survival outcomes. These valuable insights enable clinicians to tailor treatment plans, addressing a primary concern in oncology: understanding survival prospects alongside tumor traits. The second part of this study shifted to heart failure, a common cardiovascular disease with growing prevalence due to the aging population. Treatment for this condition is largely centered on drug therapy, though adherence to these therapies is often low among heart failure patients, which results in worsened health outcomes and increased hospitalizations. As a result, this pathology has a significant impact from both an epidemiological and economic standpoint. Using administrative data from the Lombardy region, this part aimed to identify risk factors and stratify patients into interpretable, survival-guided risk levels. Here, the views were artificially designed to capture diverse facets of patient management, providing a comprehensive overview of the patient’s clinical profile. This approach supports the development of precision policies, offering clinicians a structured framework to assess and manage patient risk more effectively at the population level.File | Dimensione | Formato | |
---|---|---|---|
2024_12_Iapaolo_Tesi.pdf
accessibile in internet per tutti a partire dal 19/11/2025
Descrizione: Testo della tesi
Dimensione
4 MB
Formato
Adobe PDF
|
4 MB | Adobe PDF | Visualizza/Apri |
2024_12_Iapaolo_Executive Summary.pdf
accessibile in internet per tutti a partire dal 19/11/2025
Descrizione: Executive Summary
Dimensione
583.55 kB
Formato
Adobe PDF
|
583.55 kB | 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/10589/230560