WoS İndeksli Yayınlar Koleksiyonu / WoS Indexed Publications Collection
Permanent URI for this collectionhttps://hdl.handle.net/11147/7150
Browse
2 results
Search Results
Article Novel Neural Style Transfer Based Data Synthesis Method for Phase-Contrast Wound Healing Assay Images(Elsevier Ltd, 2024) Erdem,Y.S.; Iheme,L.O.; Uçar,M.; Özuysal,Ö.Y.; Balıkçı,M.; Morani,K.; Ünay,D.; 01. Izmir Institute of TechnologyRecent advancements in the field of image synthesis have led to the development of Neural Style Transfer (NST) and Generative Adversarial Networks (GANs) which have proven to be powerful tools for data augmentation and realistic data generation. While GANs have been widely used for both data augmentation and generation, NST has not been employed for data generation tasks. Nonetheless, the simpler structure of NST compared to GANs makes it a promising alternative. In this research, we introduce an NST-based method for data generation, which to the best of our knowledge, is the first of its kind. By taking advantage of simplified architecture of NST models attributed to the utilization of a real image as the style input, our method enhances performance in data generation tasks under limited resource conditions. Additionally by utilizing patch-based training and high-resolution inference process high quality images are synthesized with limited resources. Furthermore multi-model and noised input is utilized for increased diversity with the novel NST-based data generation approach. Our proposed method utilizes binary segmentation maps as the condition input, representing the cell and wound regions. We evaluate the performance of our proposed NST-based method and compare it with a modified and fine-tuned conditional GAN (C-GAN) methods for the purpose of conditional generation of phase-contrast wound healing assay images. Through a series of quantitative and qualitative analyses, we demonstrate that our NST-based method outperforms the modified C-GAN while utilizing fewer resources. Additionally, we show that our NST-based method enhances segmentation performance when used as a data augmentation method. Our findings provide compelling evidence regarding the potential of NST for data generation tasks and its superiority over traditional GAN-based methods. The NST for data generation method was implemented in Python language and will be accessible at https://github.com/IDU-CVLab/NST_for_Gen under the MIT licence. © 2024 Elsevier LtdArticle Citation - WoS: 3Citation - Scopus: 6Improved Cell Segmentation Using Deep Learning in Label-Free Optical Microscopy Images(TÜBİTAK - Türkiye Bilimsel ve Teknolojik Araştırma Kurumu, 2021) Ayanzadeh, Aydın; Yalçın Özuysal, Özden; Pesen Okvur, Devrim; Önal, Sevgi; Yalçın Özuysal, Özden; Ünay, Devrim; 04.03. Department of Molecular Biology and Genetics; 04. Faculty of Science; 01. Izmir Institute of TechnologyThe recently popular deep neural networks (DNNs) have a significant effect on the improvement of segmentation accuracy from various perspectives, including robustness and completeness in comparison to conventional methods. We determined that the naive U-Net has some lacks in specific perspectives and there is high potential for further enhancements on the model. Therefore, we employed some modifications in different folds of the U-Net to overcome this problem. Based on the probable opportunity for improvement, we develop a novel architecture by using an alternative feature extractor in the encoder of U-Net and replacing the plain blocks with residual blocks in the decoder. This alteration makes the model superconvergent yielding improved performance results on two challenging optical microscopy image series: a phase-contrast dataset of our own (MDA-MB-231) and a brightfield dataset from a well-known challenge (DSB2018). We utilized the U-Net with pretrained ResNet-18 as the encoder for the segmentation task. Hence, following the modifications, we redesign a novel skip-connection to reduce the semantic gap between the encoder and the decoder. The proposed skip-connection increases the accuracy of the model on both datasets. The proposed segmentation approach results in Jaccard Index values of 85.0% and 89.2% on the DSB2018 and MDA-MB-231 datasets, respectively. The results reveal that our method achieves competitive results compared to the state-of-the-art approaches and surpasses the performance of baseline approaches.