Abstract
Neonatal Hypoxic-Ischemic Encephalopathy (HIE) is a leading cause of neonatal mortality and lasting neurological disability. Accurate lesion segmentation on diffusion MRI supports prognosis and treatment planning, but automated segmentation remains challenging because lesions are small, diffuse and multifocal. Moreover, apparent diffusion coefficient (ADC) voxel values, which are needed for precise segmentation, are not comparable across scanners and cohorts. A normative spatial prior, quantifying each voxel’s deviation from healthy tissue, has previously been shown to be valuable for separating injury from ordinary regional variation in ADC, though conventional approaches require a separately scanned healthy population for every new cohort, which is rarely feasible. To enable efficient transfer of a pre-trained segmentation model across cohorts, we developed a method to construct a pseudonormative spatial prior from clinical cohort data with HIE lesions, without requiring manual annotations. To further lower the labeling burden for model fine-tuning in new cohorts, we introduce a weak-supervision scheme to fine-tune the pre-trained model. HIE lesion segmentation using the pseudo-normative prior together with weak supervision performed comparably to the fully supervised models, whereas applying the pretrained model without this prior would have required scanning a separate normative population in the cohort. Model performance also surpassed the cohort’s inter-rater agreement between an experienced and an inexperienced rater. A pseudo-normative prior combined with weak supervision therefore offers a scalable route to HIE lesion segmentation in clinical practice.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_015.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{vanBen_Crosscohort_MICCAISAT2026,
author = { van Altena, Benjamin AND van der Aa, Niek E. AND de Haan, Timo R. AND Alderliesten, Thomas AND Benders, Manon J. N. L. AND Išgum, Ivana AND Venkatraghavan, Vikram},
title = { { Cross-cohort Transfer for Label-efficient Neonatal HIE Lesion Segmentation with Weak Supervision and Pseudo-normative Spatial Priors } },
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}
}