Exploring Generalization Capability of U-net Architecture through Domain Adaptation

DOI

10.2991/978-94-6463-512-6_42How to use a DOI?

Keywords

Machine Learning; Deep Learning; U-net; CNN; Medical Image Segmentation; Brain Tumor

Abstract

In today’s practice of medicine, imaging has significantly transformed the process of disease diagnosis, especially in neurology with brain tumor identification. Image segmentation, crucial for accuracy and efficiency, has been enhanced by U-net architecture. This neural network effectively segments brain tumors by extracting features and precisely localizing boundaries. Assessing its generalization abilities opens avenues for improved diagnostic methods and treatment plans, showcasing the potential of deep learning in advancing medical image analysis. The study utilized BRATS2020 from the source domain and BRATS2018 from the target domain, benchmark datasets for brain tumor identification. Data preprocessing aligns 3D NIfTI images with masks, extracts 2D slices, augments data and Volumetric data is reshaped for segmentation, making labels adjusted for accurate tumor localization. Domain adaptation for the U-net model involved retraining convolutional layers on a new dataset. Performance metrics were compared between source and target domains, showcasing adaptability and generalization. In the source domain, dice coefficient and Intersection over Union (IoU) were 0.6649 and 0.6366, respectively, demonstrating strong segmentation accuracy. Transfer to the target domain showed slightly lower metrics at 0.5788 and 0.5738. This highlights the model's generalization capability across different domains. The study underscores U-net's versatility, suggesting potential breakthroughs in domain adaptation. Further research on enhancing generalizability and internal parameters of the U-net model could advance segmentation tasks and optimize deep learning applications.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Yuhao Shen
PY  - 2024
DA  - 2024/09/23
TI  - Exploring Generalization Capability of U-net Architecture through Domain Adaptation
BT  - Proceedings of the 2024 International Conference on Artificial Intelligence and Communication (ICAIC 2024)
PB  - Atlantis Press
SP  - 393
EP  - 401
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6463-512-6_42
DO  - 10.2991/978-94-6463-512-6_42
ID  - Shen2024
ER  -