This PhD thesis aims to investigate disease-specific processing solutions in both clinical and preclinical fields, through customization, optimization and development of quantitative tools for multimodal neuroimaging analysis, adapted for each specific clinical/research question. For each specific scope, a 4 steps analysis scheme was applied and one or more steps were elaborated. 1) Techniques and contrasts that best capture anatomical and functional brain information were selected. 2) Acquisition parameters—especially for new advanced modalities—were tuned to improve contrast and signal-to-noise ratio for studying complex biological processes. 3) Post-processing pipelines were developed to combine modalities and to address for example registration, normalization, and segmentation processes. 4) Quantitative analyses, assessments and correlations were performed to allow clinicians deeper and more precise interpretations. Three main application fields were considered. 1) Cognitive Neuroscience. The specific purpose was to build models for prodromal cognitive impairment (MCI and SCD) pattern understanding and for progression to Alzheimer’s disease prediction, exploiting information obtained from the joint analysis of functional multimodal neuroimaging. Metabolic and perfusion information were combined selecting 18F-FDG PET and pCASL MR sequence (with structural MRI as processing reference), respectively. A ROI-based joint analysis pipeline was validated and applied to obtain SUVr and CBFr values. The combined ability of the two techniques to discriminate amyloid positivity (A+) in MCI/SCD subjects was quantified and their additional prognostic value for the identification of patients most likely to benefit from antibody therapy was assessed. Unsupervised cluster analysis based on a combined database demonstrated superior performance compared to analyses using a single modality. Conversely, for patients with a slight hypometabolic pattern not conclusive for neurodegenerative disease, perfusion information could discriminate A+/A- with an accuracy of 85%, showing high performances when considering ROIs of typical relevance for AD. 2) Brain Tumor (Glioblastoma - GBM). Currently, the standard of care is the total safe resection followed by radio-chemotherapy, but prognosis remains poor due to high rate of recurrence. The underlying clinical question regards the need of improvements in tumor margins definition, especially to reduce the inter-operator variability, and a deeper investigation of the peritumoral zone (PTZ), to discriminate the infiltrative tumor from the edema tissue. Firstly, a 5-point quality scale was proposed to evaluate the ability of automatic segmentations tools to correctly include/exclude tissues, based on multiple MRI sequences (T1-weigthed, T2-weighted, T2-FLAIR and contrast-enhanced T1-weigthed) acquired in standard protocols. Then, an unsupervised cluster analysis of T1ce and T2-FLAIR radiomics features was conducted in the PTZ segmentation. After qualitatively locating the first recurrence on the pre-operative scan, a preliminary qualitative assessment showed that the algorithm tended to classify potentially infiltrated tumor regions within the same cluster in simpler tumor morphologies. This results should be validated on a bigger cohort, considering features from more modalities and correlating results with histology outcomes. 3) Preclinical DWI-MRI. DWI-EPI images can give important insight into tissues microenvironment, helping in understanding specific pathological dynamics, but still needs to be optimized at preclinical level, given the high variability among vendors. A phantom-based optimization process was conducted on a Mediso nanoScan®7T relying on DTI metrics, SNR maximization and tractography quality. The process mostly involved the tuning of diffusion gradient parameters, number of shots, number of excitations and partial Fourier application

La tesi indaga soluzioni di elaborazione delle neuroimmagini patologico-specifiche (ambito clinico/preclinico), con personalizzazione, ottimizzazione e sviluppo di strumenti quantitativi per analisi multimodale, calibrati sulle esigenze di ciascuna domanda clinica/di ricerca. Per ogni ambito è stato applicato uno schema d’analisi a 4 fasi e una o più di queste fasi sono state sviluppate: 1)Selezione di tecniche che meglio catturano le informazioni anatomiche e funzionali; 2)Tuning di parametri d’acquisizione—soprattutto per nuove modalità avanzate—per migliorare qualità e rapporto segnale/rumore nello studio di processi biologici complessi; 3)Sviluppo di pipeline di post processing per combinare tecniche e gestire registrazione, normalizzazione e segmentazione; 4)Analisi quantitative, valutazioni e correlazioni per consentire interpretazioni cliniche più approfondite. Sono stati considerati tre principali ambiti applicativi. 1)Neuroscienze cognitive. Lo scopo specifico è stato costruire modelli per comprendere pattern di decadimento cognitivo prodromico (MCI/SCD) e prevedere la progressione in Alzheimer, sfruttando informazioni derivate dall’analisi congiunta di neuroimmagini multimodali funzionali. Sono state combinate informazioni metaboliche e di perfusione selezionando rispettivamente la PET 18FFDG e la sequenza MR pCASL, con MRI strutturale come riferimento di elaborazione. È stata validata e applicata una pipeline d’analisi congiunta basata su regioni di interesse per ottenere valori di SUVr e CBFr. È stata quantificata la capacità combinata delle due tecniche di discriminare la positività all’amiloide (A+) in soggetti MCI/SCD e valutato il valore prognostico aggiuntivo per identificare i pazienti più probabilmente beneficiari di terapie con anticorpi. Un database combinato usato per analisi di clustering non supervisionato ha mostrato prestazioni migliori rispetto all’uso di una singola modalità. Al contrario, per pazienti con lieve pattern ipometabolico non conclusivo per neurodegenerazione, le informazioni di perfusione hanno discriminato A+/A con accuratezza dell’85%, con elevate prestazioni considerando ROI tipiche per AD. 2)Tumore cerebrale(Glioblastoma). Lo standard attuale è la resezione totale sicura seguita da radio chemioterapia, ma la prognosi resta sfavorevole per l’alto tasso di recidiva. La questione clinica riguarda il miglioramento della definizione dei margini tumorali—in particolare per ridurre la variabilità inter operatore—e l’analisi approfondita della zona peritumorale(PTZ) per discriminare tessuto infiltrativo da edema. È stata proposta una scala di qualità a 5 punti per valutare l’accuratezza degli strumenti di segmentazione automatica, basati su più sequenze MRI (T1w,T2w,FLAIR e T1ce) acquisite in protocolli standard. È stata quindi condotta un’analisi di clustering non supervisionato nella segmentazione della PTZ, basata su caratteristiche radiomiche in T1ce e FLAIR. Dopo la localizzazione qualitativa della prima recidiva sull’immagine preoperatoria, una valutazione preliminare ha mostrato che l’algoritmo classifica i possibili tessuti infiltrati nello stesso cluster in morfologie tumorali più semplici. I risultati andranno validati su una coorte più ampia, integrando caratteristiche da più tecniche e correlando con esiti istologici. 3)DWI MRI preclinica. Le immagini DWI-EPI offrono informazioni sul microambiente tissutale, utili per comprendere dinamiche patologiche specifiche, ma richiedono ottimizzazione, soprattutto a livello preclinico, per l’elevata variabilità tra venditori. È stato condotto un processo di ottimizzazione su un Mediso nanoScan®7T, basato su metriche DTI, qualità dell’immagine in termini di massimizzazione SNR e qualità della trattografia. Il processo ha riguardato principalmente la taratura dei parametri dei gradienti di diffusione, n°di shots, n°di eccitazioni e applicazione della partial Fourier

Cerina, V (2026). Development and Optimization of Disease-Specific Tools for Multimodal Neuroimaging Quantitative Analysis. (Tesi di dottorato, , 2026).

Development and Optimization of Disease-Specific Tools for Multimodal Neuroimaging Quantitative Analysis

CERINA, VALERIA
2026

Abstract

This PhD thesis aims to investigate disease-specific processing solutions in both clinical and preclinical fields, through customization, optimization and development of quantitative tools for multimodal neuroimaging analysis, adapted for each specific clinical/research question. For each specific scope, a 4 steps analysis scheme was applied and one or more steps were elaborated. 1) Techniques and contrasts that best capture anatomical and functional brain information were selected. 2) Acquisition parameters—especially for new advanced modalities—were tuned to improve contrast and signal-to-noise ratio for studying complex biological processes. 3) Post-processing pipelines were developed to combine modalities and to address for example registration, normalization, and segmentation processes. 4) Quantitative analyses, assessments and correlations were performed to allow clinicians deeper and more precise interpretations. Three main application fields were considered. 1) Cognitive Neuroscience. The specific purpose was to build models for prodromal cognitive impairment (MCI and SCD) pattern understanding and for progression to Alzheimer’s disease prediction, exploiting information obtained from the joint analysis of functional multimodal neuroimaging. Metabolic and perfusion information were combined selecting 18F-FDG PET and pCASL MR sequence (with structural MRI as processing reference), respectively. A ROI-based joint analysis pipeline was validated and applied to obtain SUVr and CBFr values. The combined ability of the two techniques to discriminate amyloid positivity (A+) in MCI/SCD subjects was quantified and their additional prognostic value for the identification of patients most likely to benefit from antibody therapy was assessed. Unsupervised cluster analysis based on a combined database demonstrated superior performance compared to analyses using a single modality. Conversely, for patients with a slight hypometabolic pattern not conclusive for neurodegenerative disease, perfusion information could discriminate A+/A- with an accuracy of 85%, showing high performances when considering ROIs of typical relevance for AD. 2) Brain Tumor (Glioblastoma - GBM). Currently, the standard of care is the total safe resection followed by radio-chemotherapy, but prognosis remains poor due to high rate of recurrence. The underlying clinical question regards the need of improvements in tumor margins definition, especially to reduce the inter-operator variability, and a deeper investigation of the peritumoral zone (PTZ), to discriminate the infiltrative tumor from the edema tissue. Firstly, a 5-point quality scale was proposed to evaluate the ability of automatic segmentations tools to correctly include/exclude tissues, based on multiple MRI sequences (T1-weigthed, T2-weighted, T2-FLAIR and contrast-enhanced T1-weigthed) acquired in standard protocols. Then, an unsupervised cluster analysis of T1ce and T2-FLAIR radiomics features was conducted in the PTZ segmentation. After qualitatively locating the first recurrence on the pre-operative scan, a preliminary qualitative assessment showed that the algorithm tended to classify potentially infiltrated tumor regions within the same cluster in simpler tumor morphologies. This results should be validated on a bigger cohort, considering features from more modalities and correlating results with histology outcomes. 3) Preclinical DWI-MRI. DWI-EPI images can give important insight into tissues microenvironment, helping in understanding specific pathological dynamics, but still needs to be optimized at preclinical level, given the high variability among vendors. A phantom-based optimization process was conducted on a Mediso nanoScan®7T relying on DTI metrics, SNR maximization and tractography quality. The process mostly involved the tuning of diffusion gradient parameters, number of shots, number of excitations and partial Fourier application
MORESCO, ROSA MARIA
BASSO, GIANPAOLO
Neuroimmagini; Fase pre-demenza AD; Glioblastoma; ottimizzazione DWI; Analisi multimodale
Neuroimaging; Pre-dementia - AD; Glioblastoma; DWI MR optimization; Multimodal analysis
English
16-feb-2026
38
2024/2025
embargoed_20290216
Cerina, V (2026). Development and Optimization of Disease-Specific Tools for Multimodal Neuroimaging Quantitative Analysis. (Tesi di dottorato, , 2026).
File in questo prodotto:
File Dimensione Formato  
phd_unimib_897289.pdf

embargo fino al 16/02/2029

Descrizione: Development and Optimization of Disease-Specific Tools for Multimodal Neuroimaging Quantitative Analysis
Tipologia di allegato: Doctoral thesis
Dimensione 11.47 MB
Formato Adobe PDF
11.47 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/610732
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact