Πλοήγηση ανά Συγγραφέας "Lemonakis, Georgios"
Τώρα δείχνει 1 - 1 of 1
Αποτελέσματα ανά σελίδα
Επιλογές ταξινόμησης
Τεκμήριο Liver tumor segmentation using deep learning(ΕΛΜΕΠΑ, Σχολή Μηχανικών (ΣΜΗΧ), ΠΜΣ Μηχανικών Πληροφορικής, 2025-10-07) Lemonakis, Georgios; Λεμονάκης, Γεώργιος; Karampidis, Konstantinos; Καραμπίδης, ΚωνσταντίνοςLiver is a very important organ of the human body, responsible for many crucial functions. Since liver cancer is among the ones that cause many deaths worldwide, and metastases frequently have liver origin, it is essential to have high quality and accurate liver and tumor segmentations for early cancer diagnosis. From the other hand, manual identification and segmentation of lesions in three-dimensional CT scans requires too much time, is difficult to reproduce, and the segmentation results depend on the operator. To overcome these problems, many Deep Learning models have been utilized with promising results. This work is a Deep Learning approach for liver and tumor segmentation, more specifically two U-net (convolutional neural network (CNN)) variants, namely the Resnet34 and Resnet50 models, both pre-trained and afterwards fine-tuned (transfer learning), and tested on 131 CT scans of the LiTS (Liver Tumor Segmentation) challenge dataset. A 10% of these CT scans were used for evaluation of the models, and the highest DSC scores achieved were 0.926 for liver and 0.619 for tumor segmentation