Institutional Repository of Vilnius University
Duomenų augmentacijų efektyvumas nustatant ligas plaučių echoskopijose /
Abstract
dc:descriptionLung ultrasound is an attractive lung examination technique, but it is difficult to learn to interpret the images. Artificial intelligence trained with open-source POCUS datasets may help. It has been observed that AI researchers use similar augmentations to train neural network models, these augmentations are used for ordinary images. A literature research was conducted for photo augmentations that could be used to augment lung ultrasounds. Physical (deformation, reverberation, signaltonoise ratio), affine (classical), and mixing (MixUp, CutMix, HMix, GMix) augmentations were reviewed and tested. Aſter training the models with augmentations, the resulting metrics were compared to the baseline augmentation-free model to assess the effect of augmentations on the metrics. COVID class was most improved by MixUp, GMix and HMix. None of the augmentations improved the bacterial pneumonia class. GMix augmentation improves the metrics of the healthy lung class and of all classes. Physical and affine augmentations made metrics worse or did not improve enough.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Baniulis, Justas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210645012&prefLang=en_US