Development of an X-ray image processing and diagnosis tool for breast imaging using deep learning neural networks
Development of an X-ray image processing and diagnosis tool for breast imaging using deep learning neural networks
| datacite.contributor.RelatedPerson | Βασίλειος Μεγαλοοικονόμου | |
| datacite.contributor.Supervisor | Αριστείδης Δερμιτζάκης | |
| datacite.contributor.Supervisor | Νικόλαος Παλλικαράκης | |
| datacite.rights | http://purl.org/coar/access_right/c_abf2 | el |
| dc.contributor.author | Χατζάκης, Δημήτριος | |
| dc.contributor.other | Chatzakis, Dimitrios | |
| dc.date.accessioned | 2022-02-10T10:04:49Z | |
| dc.date.available | 2022-02-10T10:04:49Z | |
| dc.date.copyright | 2022-01-10 | |
| dc.degree | masterThesis | el |
| dc.description.abstract | Recently a multitude of machine learning methods and deep learning architectures have been under development for deployment in the field of medical imaging. Presented forward, a thorough analysis of breast cancer imaging analysis techniques is done with focus on the use of AI. Methods achieving results of highest accuracy of breast tumor segmentation and classification are presented with the aim of developing a robust, accurate and efficient deep learning architecture, to aid medical doctors accurately detect malignancies in an early stage. Extensive research with the inclusion of more than 350 related published papers was performed. It contained results published between 2016 to 2020 that resulted in a review of selected models, all at the vanguard of state-of-the-art progress. One easy to use, customizable and very efficient architecture, that of the UNet, was selected for benchmarking to find the optimal hyperparameter tuning and layer optimization. Two different public datasets were tested, both quantitatively and qualitatively, the MIAS and the CBIS-DDSM. The optimal results discovered for the UNet are a four-layer downsampling stream with 64 initial filters and a dropout of 0.3. Obtained results are presented in which although tumors are detected in most of the cases , other regions ,mainly in the breast’s edges, are miss detected as suspicious areas. Further work could improve on it with the addition of one or more of backbone networks in the UNet architecture. | el |
| dc.identifier.uri | http://hdl.handle.net/10889/15769 | |
| dc.language.iso | en_US | el |
| dc.subject | Breast imaging | el |
| dc.subject | Tumor segmentation | el |
| dc.subject | Tumor classification | el |
| dc.subject | Neural networks | el |
| dc.subject.alternative | Απεικόνιση μαστού | el |
| dc.subject.alternative | Ταξινόμηση νεοπλασμάτων | el |
| dc.subject.alternative | Νευρωνικά δίκτυα | el |
| dc.title | Development of an X-ray image processing and diagnosis tool for breast imaging using deep learning neural networks | el |
| dc.title.alternative | Ανάπτυξη εργαλείου ανάλυσης εικόνας και διάγνωσης για μαστογραφίες | el |
| oaire.licenseCondition | null | el |