TY - GEN
T1 - Contrastive Learning Under Noisy Temporal Self-supervision for Colonoscopy Videos
AU - Parolari, Luca
AU - Gori, Pietro
AU - Ballan, Lamberto
AU - Biffi, Carlo
AU - Le Folgoc, Loic
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2027.
PY - 2027/1/1
Y1 - 2027/1/1
N2 - Learning robust representations of polyp tracklets is key to enabling multiple AI-assisted colonoscopy applications, from polyp characterization to automated reporting and retrieval. Supervised contrastive learning is an effective approach for learning such representations, but it typically relies on correct positive and negative definitions. Collecting these labels requires linking tracklets that depict the same underlying polyp entity throughout the video, which is costly and demands specialized clinical expertise. In this work, we leverage the sequential workflow of colonoscopy procedures to derive self-supervised associations from temporal structure. Since temporally derived associations are not guaranteed to be correct, we introduce a noise-aware contrastive loss to account for noisy associations. We demonstrate the effectiveness of the learned representations across multiple downstream tasks, including polyp retrieval and re-identification, size estimation, and histology classification. Our method outperforms prior self-supervised and supervised baselines, and matches or exceeds recent foundation models across all tasks, using a lightweight encoder trained on only 27 videos. Code is available at github.com/lparolari/ntssl.
AB - Learning robust representations of polyp tracklets is key to enabling multiple AI-assisted colonoscopy applications, from polyp characterization to automated reporting and retrieval. Supervised contrastive learning is an effective approach for learning such representations, but it typically relies on correct positive and negative definitions. Collecting these labels requires linking tracklets that depict the same underlying polyp entity throughout the video, which is costly and demands specialized clinical expertise. In this work, we leverage the sequential workflow of colonoscopy procedures to derive self-supervised associations from temporal structure. Since temporally derived associations are not guaranteed to be correct, we introduce a noise-aware contrastive loss to account for noisy associations. We demonstrate the effectiveness of the learned representations across multiple downstream tasks, including polyp retrieval and re-identification, size estimation, and histology classification. Our method outperforms prior self-supervised and supervised baselines, and matches or exceeds recent foundation models across all tasks, using a lightweight encoder trained on only 27 videos. Code is available at github.com/lparolari/ntssl.
KW - Colonoscopy Videos
KW - Contrastive Learning
KW - Representation Learning
KW - Temporal Self-Supervision
UR - https://www.scopus.com/pages/publications/105053740127
U2 - 10.1007/978-3-032-38233-7_7
DO - 10.1007/978-3-032-38233-7_7
M3 - Conference contribution
AN - SCOPUS:105053740127
SN - 9783032382320
T3 - Lecture Notes in Computer Science
SP - 67
EP - 77
BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2026 - 29th International Conference, 2026, Proceedings
A2 - Yang, Guang
A2 - Adeli, Ehsan
A2 - de Bruijne, Marleen
A2 - de Bruijne, Marleen
A2 - Papiez, Bartlomiej W.
A2 - Speidel, Stefanie
A2 - Tiwari, Pallavi
A2 - Zheng, Guoyan
A2 - Yaqub, Mohammad
A2 - Dou, Qi
A2 - Rekik, Islem
PB - Springer Science and Business Media Deutschland GmbH
T2 - 29th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2026
Y2 - 27 September 2026 through 1 October 2026
ER -