Functional ultrasound is increasingly used in intra-operative, neonatal, and mobile brain imaging, but residual tissue motion, phase instability, and aliased flow can still generate artifactual high scores after conventional clutter suppression. We study localized matched-subspace detectors applied downstream of clutter filtering in beamformed slow-time fUS data, from a fixed band-limited statistic to locally whitened variants. On a held-out SIMUS structural benchmark, changing only the final detector on the same clutter-filtered residual sharply reduces nuisance leakage: in the intra-operative setting, nuisance false-positive rate falls from 0.998 for Kasai to 0.004 for the fixed matched-subspace statistic at matched 50% target-flow recall. In a separate held-out end-to-end benchmark, the selected fixed chains yield zero observed nuisance detections on all four held-out seed-setting pairs. On an open real-IQ rat-brain dataset, the fully whitened variant improves a conservative localization-derived vessel-core versus shell endpoint in all 10 audited blocks (AUC 0.530 versus 0.504 for power Doppler; exact paired sign-flip p = 0.002). These results show that the downstream detector is a consequential design choice in post-beamformed fUS, and that localized matched-subspace scoring can reduce nuisance leakage on unchanged clutter-filtered residuals.
Arthur Shune (Tue,) studied this question.