Abstract
Pediatric demyelinating diseases like multiple sclerosis (MS) are rare, yielding small annotated MRI cohorts where supervised lesion segmentation fails and adult-trained models transfer poorly. We study 2D FLAIR segmentation on PediMS (9 patients; leave-one-patient-out) with transfer to PediDemi (13 non-MS cases). Brain-constrained synthesis improves PediMS Dice over no synthesis (e.g., copy-paste from 0.510±0.094 to 0.602±0.046, Wilcoxon p = 0.004); Patch-VAE (0.589±0.042), PatchDDPM (0.590 ± 0.046), and Cross-Age (0.606 ± 0.042) paste are similar in-domain and not significantly different from one another (p = 0.65). On PediDemi transfer, cross-age paste yields the highest point estimate (0.466 ± 0.066), whereas pasting MS lesions onto lesion-free PediDemi hosts yields a negative result. Adult pretraining on MS3SEG followed by PediMS fine-tuning with copy-paste remains competitive (0.598 ± 0.039 / 0.461 ± 0.042). We also report Lesion Evidence-And-Trust (LEAT), an exploratory per-lesion score from counterfactual erasure and TTA uncertainty (High/Medium/Low TPR 0.45/0.32/0.21; Spearman r = 0.25, not exceeding lesion size as a trust signal). Thus, synthesis reliably addresses pediatric label scarcity, cross-age paste gives the best out-of-cohort transfer, and per-lesion trust scores provide exploratory quantitative explainability.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_021.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{TanArt_DataEfficient_MICCAISAT2026,
author = { Taneja, Arti AND Omer, Hasmila A. AND Vasikarla, Shantaram},
title = { { Data-Efficient Pediatric Demyelinating Lesion Segmentation with Cross-Age Lesion Augmentation and Trustworthy Predictions } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}