The transition to digital PET/CT systems, coupled with Artificial Intelligence in image reconstruction, presents unprecedented opportunities to optimize clinical workflows. This thesis evaluates the image quality stability of a latest-generation digital PET/CT (GE Omni Legend) under reduced counting statistics, aiming to define an optimized "fast-scan" protocol for clinical brain imaging. The methodology relied on an experimental design driven by decay using a Jaszczak phantom filled with 18F-FDG. A progressive reduction in signal was simulated, from 100% to 63% of the clinical baseline. The Datasets were reconstructed using the Istituto Clinico Humanitas clinical pipeline, combining the Q.Clear algorithm with deep-learning image processing. The image quality was assessed by analyzing the Background Coefficient of Variation (CoV), Contrast Recovery Coefficient (CRC), and Contrast-to-Noise Ratio (CNR) of the cold rod inserts, which emulate the morphological boundaries of brain structures. Quantitative analysis revealed a non-linear relationship between counting statistics and image degradation. The Deep Learning algorithm effectively clamped noise variance at lower count rates. Based on these findings, a 30% reduction in acquisition time was identified as the optimal trade-off. An independent validation scan confirmed that a 7-minute acquisition produces diagnostic metrics (e.g., CRC ~ 88.1%, CNR ~ 11.8) statistically and visually consistent with the standard 10-minute baseline. In conclusion, this thesis validates a 7-minute fast-scan protocol for brain PET imaging. Operationally, this optimization streamlines the workflow, increasing throughput from 4 to 6 patients per hour, while improving fragile patient comfort and reducing motion artifacts. Maintaining high diagnostic accuracy at reduced statistics paves the way for future widespread preventive screening programs with minimal radiation exposure, crucial for the early detection of neurodegenerative diseases like Alzheimer's and Parkinson's.
La transizione ai sistemi PET/TC digitali, integrata con l'Intelligenza Artificiale nella ricostruzione delle immagini, offre opportunità uniche per ottimizzare i flussi di lavoro clinici. Questa tesi valuta la stabilità della qualità d'immagine di un PET/TC digitale di ultima generazione (GE Omni Legend) a statistica di conteggio ridotta, puntando a definire un protocollo "fast-scan" ottimizzato per l'imaging cerebrale. La metodologia ha utilizzato un disegno sperimentale guidato dal decadimento con fantoccio Jaszczak (18F-FDG), simulando una progressiva riduzione del segnale dal 100% al 63% della baseline clinica. I dati sono stati ricostruiti con la pipeline dell'Istituto Clinico Humanitas (algoritmo Q.Clear e deep-learning). La qualità dell'immagine è stata valutata analizzando il Coefficiente di Variazione del fondo (CoV), il Coefficiente di Recupero del Contrasto (CRC) e il Rapporto Contrasto-Rumore (CNR) su inserti freddi, emulando i confini morfologici delle strutture cerebrali. L'analisi quantitativa ha rivelato una relazione non lineare tra statistica di conteggio e degrado dell'immagine. Il Deep Learning ha efficacemente limitato la varianza del rumore a tassi inferiori, identificando in una riduzione del 30% del tempo di acquisizione il compromesso ottimale. Una validazione indipendente ha confermato che un'acquisizione di 7 minuti produce metriche diagnostiche (es. CRC ~ 88.1%, CNR ~ 11.8) coerenti con la baseline standard di 10 minuti. In conclusione, questa tesi valida un protocollo fast-scan di 7 minuti per PET cerebrale. Operativamente, l'ottimizzazione snellisce il workflow, portando il throughput da 4 a 6 pazienti all'ora, migliorando il comfort per i soggetti fragili e riducendo gli artefatti da movimento. Preservare l'elevata accuratezza diagnostica a basse statistiche spiana la strada a futuri screening preventivi con radiazioni minime, cruciali per diagnosi precoci di patologie neurodegenerative come Alzheimer e Parkinson.
Impact of a novel generation digital Si-PMs AI-powered system in nuclear medicine workflow: a real case study
LUCCHETTI, FABRIZIO
2025/2026
Abstract
The transition to digital PET/CT systems, coupled with Artificial Intelligence in image reconstruction, presents unprecedented opportunities to optimize clinical workflows. This thesis evaluates the image quality stability of a latest-generation digital PET/CT (GE Omni Legend) under reduced counting statistics, aiming to define an optimized "fast-scan" protocol for clinical brain imaging. The methodology relied on an experimental design driven by decay using a Jaszczak phantom filled with 18F-FDG. A progressive reduction in signal was simulated, from 100% to 63% of the clinical baseline. The Datasets were reconstructed using the Istituto Clinico Humanitas clinical pipeline, combining the Q.Clear algorithm with deep-learning image processing. The image quality was assessed by analyzing the Background Coefficient of Variation (CoV), Contrast Recovery Coefficient (CRC), and Contrast-to-Noise Ratio (CNR) of the cold rod inserts, which emulate the morphological boundaries of brain structures. Quantitative analysis revealed a non-linear relationship between counting statistics and image degradation. The Deep Learning algorithm effectively clamped noise variance at lower count rates. Based on these findings, a 30% reduction in acquisition time was identified as the optimal trade-off. An independent validation scan confirmed that a 7-minute acquisition produces diagnostic metrics (e.g., CRC ~ 88.1%, CNR ~ 11.8) statistically and visually consistent with the standard 10-minute baseline. In conclusion, this thesis validates a 7-minute fast-scan protocol for brain PET imaging. Operationally, this optimization streamlines the workflow, increasing throughput from 4 to 6 patients per hour, while improving fragile patient comfort and reducing motion artifacts. Maintaining high diagnostic accuracy at reduced statistics paves the way for future widespread preventive screening programs with minimal radiation exposure, crucial for the early detection of neurodegenerative diseases like Alzheimer's and Parkinson's.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Lucchetti_Executive Summary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
18.37 MB
Formato
Adobe PDF
|
18.37 MB | Adobe PDF | Visualizza/Apri |
|
2026_03_Lucchetti_Tesi.pdf
non accessibile
Descrizione: Testo della Tesi
Dimensione
53.62 MB
Formato
Adobe PDF
|
53.62 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/10589/253684