Rewiring Development in Brain Segmentation: Leveraging Adult Brain Priors for Enhancing Infant MRI Segmentation

TL;DR

This work introduces LODi, a framework that leverages abundant adult brain priors to improve infant brain MRI segmentation across ages 0–2 years. It employs a two-stage hierarchical network trained first on large adult datasets and then adapted to infant data via weak supervision using silver-standard Infant FreeSurfer labels, augmented by a multi-level consistency training strategy. The approach demonstrates improved accuracy, robustness to motion, and generalization on internal and external datasets, including skull-stripped data, and shows qualitative alignment with expert-derived segmentations. Publicly releasing code and training volumes, the authors emphasize reproducibility and pave the way for age-aware, generalizable brain MRI segmentation across the lifespan.

Abstract

Accurate segmentation of infant brain MRI is critical for studying early neurodevelopment and diagnosing neurological disorders. Yet, it remains a fundamental challenge due to continuously evolving anatomy of the subjects, motion artifacts, and the scarcity of high-quality labeled data. In this work, we present LODi, a novel framework that utilizes prior knowledge from an adult brain MRI segmentation model to enhance the segmentation performance of infant scans. Given the abundance of publicly available adult brain MRI data, we pre-train a segmentation model on a large adult dataset as a starting point. Through transfer learning and domain adaptation strategies, we progressively adapt the model to the 0-2 year-old population, enabling it to account for the anatomical and imaging variability typical of infant scans. The adaptation of the adult model is carried out using weakly supervised learning on infant brain scans, leveraging silver-standard ground truth labels obtained with FreeSurfer. By introducing a novel training strategy that integrates hierarchical feature refinement and multi-level consistency constraints, our method enables fast, accurate, age-adaptive segmentation, while mitigating scanner and site-specific biases. Extensive experiments on both internal and external datasets demonstrate the superiority of our approach over traditional supervised learning and domain-specific models. Our findings highlight the advantage of leveraging adult brain priors as a foundation for age-flexible neuroimaging analysis, paving the way for more reliable and generalizable brain MRI segmentation across the lifespan.

Figures (8)

• Figure 1: Infant brain MRI data: (a) Internal Datasets Flow Diagram for Inclusion and Exclusion of dataset and their distributions; (b-c) Flow Diagrams for External, and Skull-stripped datasets.
• Figure 2: Proposed hierarchical 3D convolutional network for infant brain segmentation. The model follows a two-stage bottom-up training strategy: the lower-level network (grey) learns a generalised anatomical prior from adult MRIs, while the upper-level network refines segmentation on infant brain images, adapting to age-specific anatomical variations.
• Figure 3: Performance comparison: SynthSeg billot2021synthseg, LOD-Brain svanera2024fighting, and our method LODi (trained also from scratch). (a) Results on the internal dataset are computed on the test set of $1,219$ volumes from BCP and dHCP, using Infant FreeSurfer as GT reference and grouped for brain structure. (b) Results on 30 volumes belonging to external sites (MCRIB2 and ALBERTS).
• Figure 4: Comparison on the first half of 20 volumes (vol. 1-10) on the internal test set with highest disagreement on segmentation masks among difference methods. Zoom in for better view.
• Figure 5: Comparison on the second half of 20 volumes (vol. 11-20) on the external test set with highest disagreement on segmentation masks among difference methods. Zoom in for better view.