Boosting Skull-Stripping Performance for Pediatric Brain Images

TL;DR

This work targets pediatric skull-stripping, a crucial preprocessing step hindered by rapid brain development and imaging artifacts. It introduces developmental SynthStrip (d-SynthStrip), a pediatric-focused skull-stripping model trained with synthetic images derived from pediatric label maps to enhance generalization across ages and MRI contrasts. Across newborns to toddlers, d-SynthStrip outperforms pediatric baselines and matches or exceeds several adult-tailored methods, achieving higher Dice scores and lower Hausdorff distances with fast inference. The approach provides a readily usable tool for pediatric neuroimaging pipelines and highlights the value of population-specific synthetic training data.

Abstract

Skull-stripping is the removal of background and non-brain anatomical features from brain images. While many skull-stripping tools exist, few target pediatric populations. With the emergence of multi-institutional pediatric data acquisition efforts to broaden the understanding of perinatal brain development, it is essential to develop robust and well-tested tools ready for the relevant data processing. However, the broad range of neuroanatomical variation in the developing brain, combined with additional challenges such as high motion levels, as well as shoulder and chest signal in the images, leaves many adult-specific tools ill-suited for pediatric skull-stripping. Building on an existing framework for robust and accurate skull-stripping, we propose developmental SynthStrip (d-SynthStrip), a skull-stripping model tailored to pediatric images. This framework exposes networks to highly variable images synthesized from label maps. Our model substantially outperforms pediatric baselines across scan types and age cohorts. In addition, the <1-minute runtime of our tool compares favorably to the fastest baselines. We distribute our model at https://w3id.org/synthstrip.

Figures (5)

• Figure 1: SynthStrip-based training framework. Starting with manual brain label maps, we synthesize widely variable brain images and matching ground-truth brain masks, which we then use to train the model.
• Figure 2: Synthetic training images generated from pediatric label maps. The spatial and intensity variability deliberately exceeds the range of medical images to encourage d-SynthStrip to generalize across MRI contrasts and age groups.
• Figure 3: Brain extraction accuracy in terms of Hausdorff distance and volumetric Dice overlap. Testsets listed in Table \ref{['tab:age_data']}.
• Figure 4: Representative brain masks predicted by each skull-stripping method. SSCNN and deepbet focus on T1w MRI.
• Figure 5: Proportion of absolute skull-stripping errors per voxel in a nonlinear mid-space, across all images of each testset.