TY - GEN
T1 - KAN-OSD: Dino-Based Encoder and KAN-Based Decoders with Dual Contrastive Learning for One-Shot Anatomical Landmark Detection
AU - Tian, Yinbin
AU - Wang, Ziyang
AU - Xu, Shibiao
AU - Guo, Li
N1 - This is an accepted manuscript of an article published in: 2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI) . For the purposes of open access the author/s has/ve applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript (AAM) version arising from this submission.
PY - 2026/5/20
Y1 - 2026/5/20
N2 - One-shot anatomical landmark detection seeks accurate localization from a single labeled template. To address this data scarcity issue, an effective strategy is to leverage the strong generalization of a ViT-based foundation-model encoder, particularly its out-of-domain knowledge. We propose KANOSD, which freezes the encoder and trains only a lightweight prediction head with distance-aware similarity and contrastive objectives. Conventional convolutional decoders are replaced by KolmogorovArnold Networks (KAN) to model a continuous, high-resolution similarity field conditioned jointly on foundation-model features and image coordinates. To further enhance discriminability under limited supervision, we introduce metric-aware contrastive learning, which uses multiscale and multi-view augmentations of the template image as network inputs. Contrastive learning across global and local features strengthens context consistency for each landmark, while diverse augmentation views help reduce overfitting and improve generalization. KAN-OSD maintains strong label and computational efficiency without requiring any unlabeled images, while achieving lower mean radial error and higher successful detection rates on Head and Hand X-ray benchmarks.
AB - One-shot anatomical landmark detection seeks accurate localization from a single labeled template. To address this data scarcity issue, an effective strategy is to leverage the strong generalization of a ViT-based foundation-model encoder, particularly its out-of-domain knowledge. We propose KANOSD, which freezes the encoder and trains only a lightweight prediction head with distance-aware similarity and contrastive objectives. Conventional convolutional decoders are replaced by KolmogorovArnold Networks (KAN) to model a continuous, high-resolution similarity field conditioned jointly on foundation-model features and image coordinates. To further enhance discriminability under limited supervision, we introduce metric-aware contrastive learning, which uses multiscale and multi-view augmentations of the template image as network inputs. Contrastive learning across global and local features strengthens context consistency for each landmark, while diverse augmentation views help reduce overfitting and improve generalization. KAN-OSD maintains strong label and computational efficiency without requiring any unlabeled images, while achieving lower mean radial error and higher successful detection rates on Head and Hand X-ray benchmarks.
KW - Contrastive Learning
KW - Few-shot Learning
KW - Kolmogorov-Arnold Networks
KW - Landmark Detection
UR - https://ieeexplore.ieee.org/document/11515839
UR - https://www.scopus.com/pages/publications/105041657494
U2 - 10.1109/ISBI61048.2026.11515839
DO - 10.1109/ISBI61048.2026.11515839
M3 - Conference publication
T3 - Symposium Proceedings: The International Symposium on Biomedical Imaging
BT - 2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI)
PB - IEEE
ER -