Plain wording

Measuring a capillary's width at one point is enough to track how its volume changes, so full 3D segmentation is unnecessary.

slug alt-point-measurement-estimates-vessel-volumerole hypothesis

Under weak blue-light stimulation an artery and a capillary dilated while a venule did not change.

slug artery-dilates-venule-unchanged-at-low-powerrole empiricalpanel fig7A

A capillary's radius varies substantially along its own length even at rest, so a single-point width cannot stand for its volume.

slug baseline-intra-vessel-radius-varies-24pctrole empiricalpanel fig7

Capillary dilations after blue light that activates ChR2 are larger than those after green control light.

slug blue-light-dilations-exceed-green-controlrole controlpanel fig8A

Blood-flow efficiency across the capillary network rises slightly at the peak of optogenetic stimulation.

slug capillary-efficiency-increases-4pctrole empiricalpanel fig9C

Capillaries that constrict after stimulation sit deeper in cortex than those that dilate, which cluster nearer the surface.

slug constrictions-deeper-than-dilationsrole empiricalpanel fig8E

Capillaries that dilate after ChR2 activation lie closer to labelled pyramidal neurons than ones that constrict, with no such split under control light.

slug dilations-nearer-neurons-than-constrictionsrole empiricalpanel fig8D

Every neurovascular measurement comes from one deep-learning pipeline trained on a single mouse preparation and needing a GPU.

slug dl-model-scope-single-pipelinerole scopepanel fig1 (architecture)

A deep-learning segmentation and graph pipeline can measure neurovascular coupling automatically across hundreds of connected vessels at once.

slug hypothesis-dl-pipeline-enables-network-nvcrole hypothesispanel hypothesis

Activating cortical neurons drives a coordinated network-wide vascular response that single-vessel measurements cannot reveal.

slug hypothesis-network-level-nvc-coordinationrole hypothesispanel hypothesis

During strong optogenetic stimulation highly connected vessels respond together with other highly connected vessels more than at rest.

slug network-assortativity-increases-stimulationrole empiricalpanel fig9B

The NOVAS3D segmentation model gives plausible vessel outlines on another mouse strain, on rat, and on light-sheet images, without retraining.

slug novas3d-generalizes-qualitatively-oodrole empiricalpanel app1fig12, app1fig13

NOVAS3D segments vessels in three dimensions more accurately than the ilastik classifier on the deposited two-photon test images.

slug novas3d-outperforms-ilastikrole controlpanel fig3, fig4

NOVAS3D's accuracy is measured only in Thy1-ChR2-YFP mice under one imaging protocol, and in no other strain, species or modality.

slug novas3d-single-preparation-scoperole scopepanel fig1 (architecture); methods section

The pipeline should beat ilastik at segmentation, find more vessel segments than any single time point, and recover simulated radii accurately.

slug prediction-pipeline-outperforms-baselinesrole predictionpanel prediction

Blue-light stimulation should separate dilations from constrictions by distance from neurons and shift network-wide flow measures, unlike control light.

slug prediction-pipeline-reveals-network-coordinationrole predictionpanel prediction

The pipeline's radius estimator recovers simulated vessel radii well and stays stable as image noise rises.

slug radius-estimation-r2-0p68role methodologicalpanel fig5

Aligning the volumes across time points and merging their masks nearly doubles the number of vessel segments found per field of view.

slug registration-doubles-vessel-countrole methodological

A vessel counts as responding if its radius change exceeds twice its baseline variability, a threshold varied only in an appendix.

slug responder-threshold-2sd-untestedrole scopepanel app1fig14

The pipeline is benchmarked and applied in one anaesthetised mouse preparation, with generalisation to others shown only by eye.

slug scope-pipeline-and-application-paperrole scopepanel scope

Measuring vessels one at a time cannot predict how blood flow is redistributed across the cortical microvascular network.

slug synthesis-individual-vessel-measurements-insufficientrole synthesispanel synthesis (Discussion)

The UNETR ensemble traces vessel and neuron surfaces more accurately than ilastik, which over-segments vessels.

slug unetr-outperforms-ilastik-hd95role controlpanel fig3

Vessel radius changes during stimulation vary widely, and dilating vessels lie closer to active neurons than constricting ones.

slug vessel-radius-heterogeneity-stimulationrole empiricalpanel fig6

In wild-type mice without ChR2, blue and green light produce no difference in capillary radius.

slug wt-controls-no-blue-green-differencerole controlpanel app1fig9