TY - GEN
T1 - Glomeruli Segmentation in Whole-Slide Images
T2 - 2nd International Workshop on Medical Optical Imaging and Virtual Microscopy Image Analysis, MOVI 2024, held in conjunction with 26th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2024
AU - Sánchez, Maria
AU - Sánchez, Helena
AU - Arenaza, Carlos Pérez de
AU - Ribalta, David
AU - Arrarte, Nerea
AU - Cámara, Oscar
AU - Galdran, Adrian
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
PY - 2025
Y1 - 2025
N2 - We consider the task of glomeruli segmentation from Whole-Slide Images (WSIs) of pathological kidneys. In particular, we compare the performance of two different encoder-decoder architectures for two tasks: local segmentation of patches extracted from a large WSI, and global segmentation of the entire image. Since segmenting high-resolution WSIs is extremely memory-demanding, a typical approach for this task is to break down these images offline, train a patch-wise segmentation model, and then use a sliding-window inference scheme to stitch back the resulting patch segmentations. Contrary to intuition, we observe in our experiments that a model with higher segmentation accuracy at the patch level can incur in large underperformance gaps at the WSI level, even more so when measuring performance as an instance segmentation problem. This work was carried out in the context of the Kidney Pathology Image Segmentation (KPIs) challenge, which took place jointly with MICCAI 2024, and the best patch-level model we present here ranked second in the final hidden test set of the competition. Code to reproduce our experiments is shared at github.com/agaldran/kpis.
AB - We consider the task of glomeruli segmentation from Whole-Slide Images (WSIs) of pathological kidneys. In particular, we compare the performance of two different encoder-decoder architectures for two tasks: local segmentation of patches extracted from a large WSI, and global segmentation of the entire image. Since segmenting high-resolution WSIs is extremely memory-demanding, a typical approach for this task is to break down these images offline, train a patch-wise segmentation model, and then use a sliding-window inference scheme to stitch back the resulting patch segmentations. Contrary to intuition, we observe in our experiments that a model with higher segmentation accuracy at the patch level can incur in large underperformance gaps at the WSI level, even more so when measuring performance as an instance segmentation problem. This work was carried out in the context of the Kidney Pathology Image Segmentation (KPIs) challenge, which took place jointly with MICCAI 2024, and the best patch-level model we present here ranked second in the final hidden test set of the competition. Code to reproduce our experiments is shared at github.com/agaldran/kpis.
KW - Kidney Pathology Image segmentation
KW - Whole Slide Image Segmentation
UR - https://www.scopus.com/pages/publications/85218232016
U2 - 10.1007/978-3-031-77786-8_21
DO - 10.1007/978-3-031-77786-8_21
M3 - Conference contribution
AN - SCOPUS:85218232016
SN - 9783031777851
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 210
EP - 217
BT - Medical Optical Imaging and Virtual Microscopy Image Analysis - Second International Workshop, MOVI 2024, Held in Conjunction with MICCAI 2024, Proceedings
A2 - Huo, Yuankai
A2 - Millis, Bryan A.
A2 - Zhou, Yuyin
A2 - Younis, Khaled
A2 - Wang, Xiao
A2 - Tang, Yucheng
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 10 October 2024 through 10 October 2024
ER -