In the pharmaceutical and biotechnology sectors, the interaction between a drug product and its primary container, such as glass vials, plays a critical role, as the stability and integrity of the formulation must be rigorously ensured. In this context, the use of ready-to-fill (RTF) containers represents a significant added value in pharmaceutical manufacturing processes. Washing, depyrogenation, and sterilization processes subject the primary container to thermal, mechanical, and chemical stresses that may reduce its compatibility with the drug product. To date, systematic studies and guidelines capable of quantitatively correlating product stability and the surface properties of primary containers with the sterilization process are still lacking. Following the different thermal treatments, the vials were characterized to investigate their surface properties; consequently, experimental data and microscopy images were evaluated through an integrated analysis approach. Machine learning algorithms, including XGBoost, Support Vector Machines, and Logistic Regression, were employed to model the relationships between process variables and surface characteristics. Images acquired by Atomic Force Microscopy (AFM) were analyzed using Convolutional Neural Networks (CNNs) for surface classification. In contrast, images obtained by Scanning Electron Microscopy (SEM) were processed through a pipeline based on zero-shot segmentation using the Segment Anything Model (SAM) and clustering of the identified regions, leveraging both traditional morphological features and features extracted using DINOv2. The obtained results enabled a quantitative assessment of the most suitable process parameters as a function of the material type and pharmaceutical formulation.
Nel settore farmaceutico e biotecnologico, l’interazione tra il farmaco e il suo contenitore primario, come ad esempio i flaconi in vetro, riveste un ruolo critico in quanto la stabilità e l’integrità del formulato devono essere rigorosamente garantite. In questo contesto, l’impiego di contenitori ready-to-fill (RTF) rappresenta un significativo valore aggiunto nei processi di produzione dei farmaci. I processi di lavaggio, depirogenazione e sterilizzazione sottopongono il contenitore primario a sollecitazioni termiche, meccaniche e chimiche che possono ridurne la compatibilità con il farmaco. Ad oggi, mancano studi sistematici e linee guida in grado di correlare in modo quantitativo la stabilità del prodotto e le proprietà superficiali dei contenitori primari con il processo di sterilizzazione. Al termine dei diversi trattamenti termici, i flaconi sono stati caratterizzati per studiarne le proprietà superficiali, pertanto i dati sperimentali e le immagini di microscopia sono stati valutati mediante analisi integrata. Algoritmi di apprendimento automatico, tra cui XGBoost, Support Vector Machine e Logistic Regression, sono stati utilizzati per modellare le relazioni tra le variabili di processo e le caratteristiche superficiali. Le immagini acquisite mediante microscopia a forza atomica (Atomic Force Microscopy, AFM) sono state analizzate tramite reti neurali convoluzionali (Convolutional Neural Networks, CNN) per la classificazione delle superfici. Diversamente, le immagini ottenute mediante microscopia elettronica a scansione (Scanning Electron Microscopy, SEM) sono state elaborate attraverso una pipeline basata su segmentazione zero-shot con Segment Anything Model (SAM) e clustering delle regioni individuate, utilizzando sia feature morfologiche tradizionali sia feature estratte tramite DINOv2. I risultati ottenuti hanno consentito una valutazione quantitativa dei parametri di processo più idonei in funzione della tipologia di materiale e della formulazione farmaceutica.
Metodologie di Machine Learning e Deep Learning per l’analisi superficiale di flaconi in vetro mediante dati multimodali
TINELLI, ANDREA
2024/2025
Abstract
In the pharmaceutical and biotechnology sectors, the interaction between a drug product and its primary container, such as glass vials, plays a critical role, as the stability and integrity of the formulation must be rigorously ensured. In this context, the use of ready-to-fill (RTF) containers represents a significant added value in pharmaceutical manufacturing processes. Washing, depyrogenation, and sterilization processes subject the primary container to thermal, mechanical, and chemical stresses that may reduce its compatibility with the drug product. To date, systematic studies and guidelines capable of quantitatively correlating product stability and the surface properties of primary containers with the sterilization process are still lacking. Following the different thermal treatments, the vials were characterized to investigate their surface properties; consequently, experimental data and microscopy images were evaluated through an integrated analysis approach. Machine learning algorithms, including XGBoost, Support Vector Machines, and Logistic Regression, were employed to model the relationships between process variables and surface characteristics. Images acquired by Atomic Force Microscopy (AFM) were analyzed using Convolutional Neural Networks (CNNs) for surface classification. In contrast, images obtained by Scanning Electron Microscopy (SEM) were processed through a pipeline based on zero-shot segmentation using the Segment Anything Model (SAM) and clustering of the identified regions, leveraging both traditional morphological features and features extracted using DINOv2. The obtained results enabled a quantitative assessment of the most suitable process parameters as a function of the material type and pharmaceutical formulation.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi Andrea Tinelli.pdf
accesso aperto
Dimensione
6.23 MB
Formato
Adobe PDF
|
6.23 MB | Adobe PDF | Visualizza/Apri |
È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/32942