Bariatric surgery is a rapidly developing field, constantly seeking new, more efficient, and less invasive procedures. In this context, medical image segmentation, such as MRI scans, also plays an important role in further personalizing procedures to the patient’s anatomy, enabling the best solution to be designed in advance for each case. Currently, segmentation is performed by medical professionals, but the long-term goal is to automate the process. Automation would not only reduce time and costs but also minimize issues related to human error. Convolutional neural networks (CNNs) represent the future of medicine, although, at present, they are still considered somewhat unripe as tools. To achieve a fully automatic process, CNNs require extensive training, which depends on a large dataset to draw from. The objective of this work is to train neural networks by providing MRI segmentation examples. These examples will then be used by the network to create a database that will ultimately enable it to autonomously segment additional medical images. Specifically, this study will analyze two groups of patients undergoing two of the most common bariatric techniques: laparoscopic sleeve gastrectomy (LSG) and endoscopic sleeve gastroplasty (ESG). The segmentation of preoperative MRIs will provide data on stomach volumes and gastric emptying rates, which will later be used in statistical analyses to explore factors that may correlate with obesity and influence it.
La chirurgia bariatrica è un settore in ampio sviluppo, in costante ricerca di nuove procedure, più efficienti e meno invasive. In tale contesto si inserisce correttamente anche la segmentazione di immagini mediche, quali risonanze magnetiche (MRI), proprio per personalizzare ulteriormente l’intervento all’anatomia del paziente, e dunque progettare in anticipo la miglior soluzione del caso. Il lavoro di segmentazione tuttora è eseguito dai medici, ma l’obiettivo a lungo termine è di automatizzare il processo, così da ridurre le tempistiche e i costi, ma soprattutto le possibili problematiche dovute all’incertezza dell’operato umano. Le reti CNN rappresentano quindi il futuro della medicina, anche se ora come ora sono da considerarsi ancora uno strumento acerbo, in quanto per poter davvero ottenere un processo completamente automatico hanno bisogno di addestramento e dunque di un’ampia raccolta di dati da cui attingere. L’addestramento delle reti neurali è proprio l’obiettivo di questo lavoro, fornire cioè degli esempi di segmentazione di MRI, i quali verranno poi utilizzati dalla rete per creare un database da cui ricavare gli strumenti per segmentare autonomamente ulteriori immagini mediche. In particolare, verranno analizzate due coorti di pazienti soggette a due delle tecniche più diffuse di gastroplastica: la gastroplastica verticale laparoscopica (LSG) e la gastroplastica verticale endoscopica (ESG). Il lavoro di segmentazione delle MRI preoperatorie dei pazienti fornirà poi dati riguardanti i volumi gastrici e le velocità di svuotamento dello stomaco, i quali saranno infine utilizzati all’interno di un’analisi statistica per indagare quali altri fattori possono essere correlati al fenomeno dell’obesità e come la influenzano.
Reti neurali per la chirurgia bariatrica: annotazione di immagini del distretto gastrico in fase pre-operatoria per lo sviluppo di un database ground truth
BODINI, ENNIA
2023/2024
Abstract
Bariatric surgery is a rapidly developing field, constantly seeking new, more efficient, and less invasive procedures. In this context, medical image segmentation, such as MRI scans, also plays an important role in further personalizing procedures to the patient’s anatomy, enabling the best solution to be designed in advance for each case. Currently, segmentation is performed by medical professionals, but the long-term goal is to automate the process. Automation would not only reduce time and costs but also minimize issues related to human error. Convolutional neural networks (CNNs) represent the future of medicine, although, at present, they are still considered somewhat unripe as tools. To achieve a fully automatic process, CNNs require extensive training, which depends on a large dataset to draw from. The objective of this work is to train neural networks by providing MRI segmentation examples. These examples will then be used by the network to create a database that will ultimately enable it to autonomously segment additional medical images. Specifically, this study will analyze two groups of patients undergoing two of the most common bariatric techniques: laparoscopic sleeve gastrectomy (LSG) and endoscopic sleeve gastroplasty (ESG). The segmentation of preoperative MRIs will provide data on stomach volumes and gastric emptying rates, which will later be used in statistical analyses to explore factors that may correlate with obesity and influence it.File | Dimensione | Formato | |
---|---|---|---|
Bodini_Ennia.pdf
embargo fino al 19/11/2027
Dimensione
4.17 MB
Formato
Adobe PDF
|
4.17 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/76466