Abstract
Carotid strain is a promising marker of arterial stiffening and subclinical atherosclerosis, measured from the change in wall perimeter across the cardiac cycle. Per-frame wall segmentation is essentially solved, but the hard property for strain is temporal consistency, a stable contour across frames, which Dice misses. Validation against one expert’s sparse annotations is doubly problematic: experts label few cycles, and a model trained on that expert imitates it, so high agreement can mislead. We therefore anchor and test against references the labels cannot provide: the ECG, and the ESC SCORE2 risk score that strain should track. On a held-out subgroup of participants from CNIC’s prospective cardiovascular cohorts (40 participants, 71 scans) with synchronised ECG, a full-field multi-frame nnU-Net reproduced the expert frame by frame (Dice , perimeter ) and recovered the heartbeat (, MAE bpm). Dense automatic strain and the expert’s sparse reading both tracked risk ( and ), and after adjusting for age and blood pressure, only dense strain stayed sig-nificant (, ). Per-frame Dice was flat ( to ) while the SCORE2 association ranged from for the single-frame model to for temporal-context models, so per-frame accuracy does not predict clinical validity but temporal consistency does. A single-seed optical-flow tracker recovered a similar association (), confirming temporal consistency as the active ingredient. When the reference is sparse, agreement is the wrong target: a consistent automatic strain, anchored to ECG and associated with SCORE2, offers a scalable cardiovascular-risk read-out from a routine carotid scan.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ASMUS_036.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=pmw7wA9Vx4
BibTex
@InProceedings{NavAlv_Beyond_MICCAISAT2026,
author = { Navarro-Guzman, Alvaro AND Mass, Virginia AND Garcia-Lunar, Ines AND Fuster, Valentin AND Sanchez-Gonzalez, Javier AND Ibañez, Borja},
title = { { Beyond Per-Frame Accuracy: Temporally Consistent Carotid Strain Tracks Cardiovascular Risk } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17276},
month = {pending},
page = {pending}
}