Abstract
Automated CTA analysis for deep inferior epigastric perforator (DIEP) flap planning requires both vessel segmentation and variable-count perforator keypoint localization. Joint optimization of these tasks is challenging because their annotations differ substantially in density and supervision strength. We propose PerfSeg, a two-stage structure-conditioned framework, to first train and freeze a vessel segmentation network and then use the resulting vascular representations to guide key-point localization. In particular, t he localization stage combines vessel and muscle–adipose interface features through Dual Spatial Cross Attention and suppresses anatomically implausible responses through Soft Gating. Experiments are conducted on 277 clinical CTA cases, which are divided into 221 training, 28 validation, and 28 held-out test cases. The selected vessel model achieves a clDice of 0.745 on the validation set. On the held-out test set, PerfSeg achieves a precision of 0.679 and a recall of 0.708 under one-to-one matching at the predefined 10-mm threshold. The mean localization error (MLE) among successfully matched pairs was 7.8 mm, while the symmetric Chamfer distance (CD) over the complete point sets is 17.6 mm. Compared with four alternative localization methods, PerfSeg achieves the highest precision and recall and the lowest CD. These results demonstrate the value of combining vascular and tissue-interface priors for anatomically constrained perforator localization.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_088.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=QIBeWLFEts
BibTex
@InProceedings{XuChi_TwoStage_MICCAISAT2026,
author = { Xu, Chi AND Li, Xuejian AND Zhang, Zhengguo AND Ouyang, Xi AND Xue, Zhong AND Shen, Dinggang},
title = { { Two-Stage Structure-Conditioned Vessel Segmentation and Keypoint Localization for DIEP Flap Planning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}