Hemi-Brain Growth

Metallic shunt valves generate ferromagnetic susceptibility artifacts on MRI, causing signal loss and geometric distortion that can render up to 30% of hemi-brain volume non-diagnostic. Crucially, such artifacts never affect the contralateral hemisphere. This study tests whether the stable relationship between left and right hemisphere volumes, well-documented in normal childhood development, persists postoperatively in hydrocephalus patients, enabling the artifact-free hemisphere to serve as a reliable longitudinal proxy for whole-brain growth.

Statistical analysis (Mann-Whitney U-Test) confirmed no significant difference between early postoperative (0–180 days) and late postoperative (>180 days) hemisphere volume ratios (U = 1326.0, p = 0.4858), validating the hemi-brain approach. A significant shift was observed between preoperative and early postoperative ratios (U = 1879.5, p = 0.0288), reflecting the acute effects of surgical intervention before stabilization.

We propose the Hemi-Brain Growth Curve as a novel proxy for assessing whole-brain growth when metallic artifacts obscure one hemisphere. The approach is grounded in the observation that left and right hemisphere volumes grow symmetrically in normal childhood development and that this ratio stabilizes following surgical intervention. The method does not assume hemispheric symmetry, only that the per-patient postoperative ratio is stable over follow-up time, making the unaffected hemisphere sufficient for longitudinal monitoring.

An AI-based brain volume estimation pipeline was developed for postoperative MRI with shunt artifact. The pipeline consists of three steps: (1) obtain brain/CSF segmentation maps using a fully convolutional Dense U-Net, (2) generate hemi-brain masks of the artifact-unaffected side using a DL-based mask generator, and (3) calculate brain and CSF volume from the segmentation maps and voxel slice thickness. The Dense U-Net leverages DenseNet-style skip-connected blocks within an encoder–decoder framework, trained with a joint Dice + masked Total Variation loss to produce smooth, accurate CSF boundary predictions.

1. U. S. Bajaj*, M. Yu*, K. Templeton*, S. Mukherjee*, H. Zhang, N. Nunn, A. V. Kulkarni, J. Kestle, V. Monga, S. J. Schiff, "Hemi-brain growth as a biomarker for whole brain growth," Journal of Neurosurgery: Pediatrics. [Code] [Paper]