Work

Voxel-wise temporal decomposition of hypoxia-powered BOLD MRI: method development and proof-of-concept application in glioblastoma

Year: 2026

Type: preprint

Abstract: Hypoxia-powered blood oxygen level-dependent (BOLD) MRI probes the BOLD response to a controlled transient hypoxic stimulus. We developed GLIODISSECT, an operational voxel-wise framework that separates relative response arrival, descending and recovery transitions, and threshold-defined steady state amplitude. Twenty healthy controls underwent 3-T BOLD MRI during a double hypoxic-step challenge, and three patients with newly diagnosed glioblastoma were included as proof-of-concept cases. Full-record cross-correlation and first-step 10% landmarks yielded one refined voxel-wise lag that was reused downstream. Alternative 5%/95% and 20%/80% thresholds preserved steady state map patterns (median within-subject spatial r 0.9987-0.9998), while timing measures were more threshold-dependent. Numerical criteria were satisfied in 97.789% and 97.682% of evaluable delay-to-plateau (DTP) estimates and in all evaluable delay-to-return (DTR) estimates. Leave-one-subject-out validation of the global-response-adjusted steady state model yielded a median voxel-wise Q 2 of 0.345 and reduced pooled prediction error by 16.0%, improving 19 of 20 controls. In the three illustrative cases, component and deviation maps displayed spatially heterogeneous timing and steady state patterns that could not be represented by one response-amplitude summary. GLIODISSECT supports component-specific mapping and voxel-wise normative interpretation. Biological and clinical validation remain required.

Source: medRxiv

DOI: 10.64898/2026.05.27.26354265