Liver ultrasound is a diagnostic examination prescribed in cases of symptoms associated with liver disease. It is commonly used due to its speed, affordability, ease of use, portability, and non-invasive nature. Real-time segmentation of liver ultrasound is essential to provide a clear clinical overview of organ morphology and conditions (such as the presence of hepatocellular carcinoma or hepatic steatosis) and to assist surgeons and radiologists in therapeutic procedures. The objective of this thesis is to study the impact of image preprocessing on the performance of a neural network trained for liver ultrasound segmentation. The network was developed in MATLAB using the DeepLabv3Plus model with a ResNet18 backbone and was trained on a publicly available dataset of 2400 ultrasound images from 8 healthy volunteers.
L’ecografia epatica `e un’esame diagnostico prescritto in caso di sintomi associati ad una malattia del fegato ed `e comunemente utilizzata in quanto caratterizzata da rapidità, basso costo, facilità d’uso, portabilità e natura non invasiva. La segmentazione in tempo reale dell’ecografia epatica `e essenziale per delineare un quadro clinico chiaro riguardo la morfologia e le condizioni dell’organo (ad esempio la presenza di un carcinoma epatocellulare o di una steatosi epatica) e per assistere i chirurghi e i radiologi nelle procedure terapeutiche. L’obiettivo di questa tesi `e quello di studiare l’impatto del preprocessing delle immagini sulle prestazioni di un rete neurale addestrata per la segmentazione di ecografie epatiche. La rete `e stata realizzata su MATLAB con il modello DeepLabv3Plus con backbone ResNet18 ed `e stata allenata con un dataset di ecografie disponibile pubblicamente contenente 2400 immagini di 8 volontari sani.
Machine learning per l'imaging biomedico: valutazione di DeepLabv3 per la segmentazione dell'ecografia epatica
GIROTTO, RICCARDO
2022/2023
Abstract
Liver ultrasound is a diagnostic examination prescribed in cases of symptoms associated with liver disease. It is commonly used due to its speed, affordability, ease of use, portability, and non-invasive nature. Real-time segmentation of liver ultrasound is essential to provide a clear clinical overview of organ morphology and conditions (such as the presence of hepatocellular carcinoma or hepatic steatosis) and to assist surgeons and radiologists in therapeutic procedures. The objective of this thesis is to study the impact of image preprocessing on the performance of a neural network trained for liver ultrasound segmentation. The network was developed in MATLAB using the DeepLabv3Plus model with a ResNet18 backbone and was trained on a publicly available dataset of 2400 ultrasound images from 8 healthy volunteers.File | Dimensione | Formato | |
---|---|---|---|
Girotto_Riccardo.pdf
accesso aperto
Dimensione
3.52 MB
Formato
Adobe PDF
|
3.52 MB | Adobe PDF | Visualizza/Apri |
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/20.500.12608/48839