The prediction of radiation-induced biological damage plays a crucial role in both the optimization of cancer treatment plans and the evaluation of potential side effects on healthy tissue. Since such biological damage is primarily caused by ionization events at the DNA scale—a macromolecule approximately 2 nm in diameter—understanding radiation interactions at nanometric dimensions is fundamental for accurately modeling biological effects. Nanodosimetry is the field of radiation physics that investigates particle interaction sites at this scale and provides key quantities for the descripion of the biological damage by measuring distributions related to ionizing events.. To date, only three nanodosimeters worldwide are capable of performing such measurements; however, their size and complexity render them unsuitable for clinical implementation. This thesis investigates the feasibility of developing a compact and portable nanodosimeter based on a segmented-anode design, derived from a Tissue-Equivalent Proportional Counter (TEPC) architecture traditionally used in microdosimetry. The study employs Monte Carlo simulations to reproduce the detector’s internal geometry, electric field distribution, and the electron avalanche multiplication process. After validation against experimental data available in the literature, the simulation model was used to evaluate the number of Townsend avalanches in the amplification region, enabling an estimation of the number of primary ionization events occurring within the sensitive volume. Three detector configurations with segmented anodes composed of 10, 15, and 20 elements, respectively, were studied. A machine learning (ML) framework was developed to infer the original ionization cluster size from the anode signal distributions. For this purpose, fully labeled datasets were generated through simulations, where each event (e.g., from 10,000 protons at 6 MeV) was represented by a vector of collected avalanche electrons per segment, along with the corresponding number of primary ionizations. The ML model’s performance was assessed in terms of its ability to reconstruct the ionization cluster size distribution. The segmentation scheme that provided the optimal trade-off between feature dimensionality and predictive accuracy was selected for further development. To physically enable the readout of signals from individual anode segments, a custom 6-layer printed circuit board (PCB) was also designed, which was later ordered for manufacturing. This PCB layout was optimized to ensure signal integrity and compatibility with the spatially segmented detection architecture, paving the way for the implementation of the proposed nanodosimetric system. This work was carried out within the framework of NECTAR, a project funded under the FET-OPEN program (Grant Agreement No. 964934).
La previsione del danno biologico indotto dalle radiazioni riveste un ruolo cruciale sia nell’ottimizzazione dei piani di trattamento oncologico, sia nella valutazione dei potenziali effetti collaterali sui tessuti sani. Poiché tale danno biologico è principalmente causato da eventi di ionizzazione alla scala del DNA—una macromolecola con un diametro di circa 2 nm—comprendere le interazioni delle radiazioni a dimensioni nanometriche è fondamentale per modellare con precisione gli effetti biologici. La nanodosimetria è il ramo della fisica delle radiazioni che studia i siti di interazione delle particelle a questa scala e fornisce grandezze fisiche chiave per descrivere il danno biologico, misurando distribuzioni associate agli eventi ionizzanti. Attualmente, solo tre nanodosimetri al mondo sono in grado di effettuare tali misurazioni; tuttavia, le loro dimensioni e complessità ne rendono impraticabile l’impiego in ambito clinico. La presente tesi esplora la fattibilità dello sviluppo di un nanodosimetro compatto e portatile, basato su un design a anodo segmentato, derivato da un’architettura di tipo Tissue-Equivalent Proportional Counter (TEPC), tradizionalmente impiegata in microdosimetria. Lo studio fa uso di simulazioni Monte Carlo per riprodurre la geometria interna del rivelatore, la distribuzione del campo elettrico e il processo di moltiplicazione a valanga degli elettroni. Dopo una fase di validazione rispetto a dati sperimentali presenti in letteratura, il modello simulativo è stato utilizzato per valutare il numero di valanghe di Townsend nella regione di amplificazione, permettendo una stima del numero di ionizzazioni primarie avvenute nel volume sensibile. Sono state analizzate tre configurazioni di rivelatore con anodi segmentati composti rispettivamente da 10, 15 e 20 elementi. È stato sviluppato un framework di machine learning (ML) per inferire la dimensione originaria del cluster di ionizzazione a partire dalla distribuzione dei segnali raccolti agli anodi. A tal fine, sono stati generati dataset completamente etichettati tramite simulazioni, in cui ciascun evento (ad esempio, 10.000 protoni a 6 MeV) era rappresentato da un vettore di elettroni raccolti per segmento, associato al corrispondente numero di ionizzazioni primarie. La performance del modello ML è stata valutata in termini di capacità di ricostruzione della distribuzione delle dimensioni dei cluster. La configurazione di segmentazione che ha fornito il miglior compromesso tra dimensionalità delle feature e accuratezza predittiva è stata selezionata per un ulteriore sviluppo. Per consentire fisicamente la lettura dei segnali provenienti dai singoli segmenti dell’anodo, è stato inoltre progettato un circuito stampato (PCB) multistrato a 6 strati, successivamente ordinato per la produzione. Il layout del PCB è stato ottimizzato per garantire l’integrità del segnale e la compatibilità con l’architettura di rilevazione segmentata, aprendo la strada all’implementazione del sistema nanodosimetrico proposto. Il lavoro è stato svolto nell’ambito del progetto NECTAR, finanziato dal programma FET-OPEN (Grant Agreement No. 964934).
Design and development of a gas nanodosimeter with segmented anode: a machine learning-based optimization approach
Gasperini, Giovanni
2024/2025
Abstract
The prediction of radiation-induced biological damage plays a crucial role in both the optimization of cancer treatment plans and the evaluation of potential side effects on healthy tissue. Since such biological damage is primarily caused by ionization events at the DNA scale—a macromolecule approximately 2 nm in diameter—understanding radiation interactions at nanometric dimensions is fundamental for accurately modeling biological effects. Nanodosimetry is the field of radiation physics that investigates particle interaction sites at this scale and provides key quantities for the descripion of the biological damage by measuring distributions related to ionizing events.. To date, only three nanodosimeters worldwide are capable of performing such measurements; however, their size and complexity render them unsuitable for clinical implementation. This thesis investigates the feasibility of developing a compact and portable nanodosimeter based on a segmented-anode design, derived from a Tissue-Equivalent Proportional Counter (TEPC) architecture traditionally used in microdosimetry. The study employs Monte Carlo simulations to reproduce the detector’s internal geometry, electric field distribution, and the electron avalanche multiplication process. After validation against experimental data available in the literature, the simulation model was used to evaluate the number of Townsend avalanches in the amplification region, enabling an estimation of the number of primary ionization events occurring within the sensitive volume. Three detector configurations with segmented anodes composed of 10, 15, and 20 elements, respectively, were studied. A machine learning (ML) framework was developed to infer the original ionization cluster size from the anode signal distributions. For this purpose, fully labeled datasets were generated through simulations, where each event (e.g., from 10,000 protons at 6 MeV) was represented by a vector of collected avalanche electrons per segment, along with the corresponding number of primary ionizations. The ML model’s performance was assessed in terms of its ability to reconstruct the ionization cluster size distribution. The segmentation scheme that provided the optimal trade-off between feature dimensionality and predictive accuracy was selected for further development. To physically enable the readout of signals from individual anode segments, a custom 6-layer printed circuit board (PCB) was also designed, which was later ordered for manufacturing. This PCB layout was optimized to ensure signal integrity and compatibility with the spatially segmented detection architecture, paving the way for the implementation of the proposed nanodosimetric system. This work was carried out within the framework of NECTAR, a project funded under the FET-OPEN program (Grant Agreement No. 964934).File | Dimensione | Formato | |
---|---|---|---|
2025_07_Gasperini_ExecutiveSummary.pdf
non accessibile
Dimensione
3.94 MB
Formato
Adobe PDF
|
3.94 MB | Adobe PDF | Visualizza/Apri |
2025_07_Gasperini_Tesi.pdf
non accessibile
Dimensione
14.5 MB
Formato
Adobe PDF
|
14.5 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/10589/239907