Yongsheng Zhang, Haojie Han, Fang Chen, Hongen Liao
Neurodevelopmental expansion of human association cortex increases cortical folding complexity, causing registration inaccuracies that confound detection of subtle pathological signatures like focal cortical dysplasia (FCD) in voxel-based morphometry. We introduce INFREG (volume-inflation registration), integrating surface-based deformation with volumetric alignment to minimize sulcal smoothing. Validated on 80 histologically confirmed FCD cases and 85 controls, INFREG outperformed conventional methods (CAT12/Demons/BrainSuite) by (i) superior template to subject alignment (SSIM = 0.95 ± 0.04; GM/WM Dice = 0.93 ± 0.05 vs. 0.70-0.87, all p < 0.001); (ii) sharper templates with 23%-41% lower intensity SD and finer gyral alignment (3.34 ± 0.13 vs. 3.45 ± 0.15 mm, p < 0.01); (iii) reducing false positive FCD Type II clusters by 70%-88% (p < 0.001) in association cortices; and (iv) halving manual verification time (Observer1:101 ± 67 vs.177 ± 114 s; Observer2:75 ± 48 vs.162 ± 102 s; p < 0.001). In 30 typical cases, 90% localized within top 5 clusters, and all 12 VBM-indeterminate cases resolved. INFREG may enhance clinical workflows for FCD type II evaluation.