TY - GEN
T1 - SwinVFTR
T2 - 22nd IEEE International Symposium on Biomedical Imaging, ISBI 2025
AU - Hossain, Khondker Fariha
AU - Kamran, Sharif Amit
AU - Tavakkoli, Alireza
AU - Bebis, George
AU - Baker, Sal
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - Accurately segmenting fluid in 3D optical coherence tomography (OCT) images is critical for detecting eye diseases but remains challenging. Traditional autoencoder-based methods struggle with resolution loss and information recovery. While transformer-based models improve segmentation, they aren't optimized for 3D OCT volumes, which vary by vendor and extraction technique. To address this, we propose SwinVFTR, a transformer architecture for precise fluid segmentation in 3D OCT images. SwinVFTR employs channel-wise volumetric sampling and a shifted window transformer block to improve fluid localization. Moreover, a novel volumetric attention block enhances spatial and depth-wise attention. Trained using multiclass dice loss, SwinVFTR outperforms existing models on Spectralis, Cirrus, and Topcon OCT datasets, achieving mean dice scores of 0.72, 0.59, and 0.68, respectively, along with superior performance in mean intersection-over-union (IOU) and structural similarity (SSIM) metrics.
AB - Accurately segmenting fluid in 3D optical coherence tomography (OCT) images is critical for detecting eye diseases but remains challenging. Traditional autoencoder-based methods struggle with resolution loss and information recovery. While transformer-based models improve segmentation, they aren't optimized for 3D OCT volumes, which vary by vendor and extraction technique. To address this, we propose SwinVFTR, a transformer architecture for precise fluid segmentation in 3D OCT images. SwinVFTR employs channel-wise volumetric sampling and a shifted window transformer block to improve fluid localization. Moreover, a novel volumetric attention block enhances spatial and depth-wise attention. Trained using multiclass dice loss, SwinVFTR outperforms existing models on Spectralis, Cirrus, and Topcon OCT datasets, achieving mean dice scores of 0.72, 0.59, and 0.68, respectively, along with superior performance in mean intersection-over-union (IOU) and structural similarity (SSIM) metrics.
KW - Fluid Segmentation
KW - OCT Segmentation
KW - Optical Coherence Tomography
KW - Swin Transformer
UR - https://www.scopus.com/pages/publications/105005827181
UR - https://www.scopus.com/pages/publications/105005827181#tab=citedBy
U2 - 10.1109/ISBI60581.2025.10980687
DO - 10.1109/ISBI60581.2025.10980687
M3 - Conference contribution
AN - SCOPUS:105005827181
T3 - Proceedings - International Symposium on Biomedical Imaging
BT - ISBI 2025 - 2025 IEEE 22nd International Symposium on Biomedical Imaging, Proceedings
PB - IEEE Computer Society
Y2 - 14 April 2025 through 17 April 2025
ER -