TY - GEN
T1 - OG-SAM
T2 - 4th International Workshop on Computational Mathematics Modeling in Cancer Analysis, CMMCA 2025, Held in Conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025
AU - Wu, Xidong
AU - Chen, Hao
AU - Li, Zhuoyuan
AU - Li, Chao
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - The Segment Anything Model (SAM) excels in image segmentation. yet is challenged in multi-organ segmentation, due to the inherent similarities between organ tissues and the substantial variability in organ size, structure, and texture. This paper proposes to guide the adaptation of SAM for multi-organ segmentation, introducing biological priors of organogenesis, where organs arise from specific germ layers and develop with shared early-stage pathways before divergence into unique structures. We present OG-SAM (Organogenesis SAM), a new paradigm that enables organ-wise adaptation. First, we present OrganAdapt (Organ Adaptation) that integrates a biologically inspired hierarchical adaptation module into SAM, where parameter sharing and specialization follow the developmental trajectory of organs. Second, to effectively address variations in organ size, we propose GoF (Generalized Organ-feature Fusion), a mechanism that facilitates organ-specific multiscale feature pyramid fusion, thereby enhancing segmentation accuracy and robustness. OG-SAM functions as a query-based plug-in, seamlessly integrating with SAM. Experiments show that OG-SAM outperforms competing methods, particularly for challenging organ boundaries.
AB - The Segment Anything Model (SAM) excels in image segmentation. yet is challenged in multi-organ segmentation, due to the inherent similarities between organ tissues and the substantial variability in organ size, structure, and texture. This paper proposes to guide the adaptation of SAM for multi-organ segmentation, introducing biological priors of organogenesis, where organs arise from specific germ layers and develop with shared early-stage pathways before divergence into unique structures. We present OG-SAM (Organogenesis SAM), a new paradigm that enables organ-wise adaptation. First, we present OrganAdapt (Organ Adaptation) that integrates a biologically inspired hierarchical adaptation module into SAM, where parameter sharing and specialization follow the developmental trajectory of organs. Second, to effectively address variations in organ size, we propose GoF (Generalized Organ-feature Fusion), a mechanism that facilitates organ-specific multiscale feature pyramid fusion, thereby enhancing segmentation accuracy and robustness. OG-SAM functions as a query-based plug-in, seamlessly integrating with SAM. Experiments show that OG-SAM outperforms competing methods, particularly for challenging organ boundaries.
KW - Multi-Organ Segmentation
KW - Organ-specific Adaptation
KW - Segment Anything Model
UR - https://www.scopus.com/pages/publications/105018799245
U2 - 10.1007/978-3-032-06624-4_9
DO - 10.1007/978-3-032-06624-4_9
M3 - Conference contribution
AN - SCOPUS:105018799245
SN - 9783032066237
VL - 16178
T3 - Lecture Notes in Computer Science
SP - 80
EP - 89
BT - Computational Mathematics Modeling in Cancer Analysis - 4th International Workshop, CMMCA 2025, Held in Conjunction with MICCAI 2025, Proceedings
A2 - Li, Chao
A2 - Qin, Wenjian
A2 - Wu, Jia
A2 - Zaki, Nazar
PB - Springer Science and Business Media Deutschland GmbH
CY - Switzerland
Y2 - 27 September 2025 through 27 September 2025
ER -