{ "type": "Document", "metadata": { "doi": "10.7554/eLife.95525", "title": "A deep learning pipeline for mapping in situ network-level neurovascular coupling in multi-photon fluorescence microscopy", "authors": [ "Rozak et al." ] }, "children": [ { "type": "ClaimGraph", "identifier": "rozak-2026-neurovascular-dl-claims", "children": [ { "type": "Claim", "identifier": "alt-point-measurement-estimates-vessel-volume", "role": "hypothesis", "children": [ { "type": "Text", "value": "Vessel caliber measured at a single point along a capillary is an adequate estimate of microvessel volume change, so point measurements suffice for quantifying neurovascular coupling without volumetric segmentation." } ], "stance": "rejects", "metadata": { "uuid": "d58f5ce7-bd4f-461e-9b01-4550aa813eff", "concepts": [ "point measurement", "vessel caliber", "alternative explanation" ] } }, { "type": "Claim", "identifier": "artery-dilates-venule-unchanged-at-low-power", "role": "empirical", "children": [ { "type": "Text", "value": "At 1.1 mW/mm² 458 nm stimulation, a sample artery dilated 1.33±0.86 µm (p<1e-4) and a sample capillary dilated 0.42±0.39 µm (p<1e-4), while a sample venule showed no significant radius change (p=0.22), demonstrating vessel-type heterogeneity in optogenetic neurovascular responses." } ], "relations": [ { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:disagreesWith" }, { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:supports" } ], "metadata": { "uuid": "af8bc967-3346-4e25-ac9e-8ad3536313fd", "concepts": [ "artery", "venule", "capillary", "vessel-type heterogeneity", "optogenetic neurovascular response" ] } }, { "type": "Claim", "identifier": "baseline-intra-vessel-radius-varies-24pct", "role": "empirical", "children": [ { "type": "Text", "value": "Capillary radius varies along vessel length by 24±28% of the mean resting radius across baseline frames, such that point measurements of vessel caliber cannot accurately estimate microvessel volume changes." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "alt-point-measurement-estimates-vessel-volume", "relationType": "claimrel:rulesOut" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "synthesis-individual-vessel-measurements-insufficient", "relationType": "cito:confirms" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:supports" } ], "metadata": { "uuid": "083ef4c3-9e93-4927-9867-3c785b2c03c0", "concepts": [ "intra-vessel radius heterogeneity", "capillary morphology", "point measurement limitation", "vessel volume" ] } }, { "type": "Claim", "identifier": "blue-light-dilations-exceed-green-control", "role": "control", "children": [ { "type": "Text", "value": "Capillary dilations following 458 nm ChR2 stimulation (0.90±0.93 µm at 1.1 mW/mm²; 0.90±0.78 µm at 4.3 mW/mm²) are significantly larger than those following 552 nm control illumination (0.58±0.92 µm; p<1e-4), confirming that the vascular responses are ChR2-mediated rather than photothermal artifacts." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:confirms" }, { "xref": "constrictions-deeper-than-dilations", "relationType": "cito:confirms" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:confirms" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "cito:confirms" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "cito:confirms" }, { "xref": "wt-controls-no-blue-green-difference", "relationType": "claimrel:requires" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:supports" } ], "metadata": { "uuid": "a68ebd2f-95bf-43a9-b0e0-f9ce7a3fa5bd", "concepts": [ "ChR2 specificity", "optogenetic control", "capillary dilation", "photothermal artifact", "neurovascular coupling" ] } }, { "type": "Claim", "identifier": "capillary-efficiency-increases-4pct", "role": "empirical", "children": [ { "type": "Text", "value": "Capillary network efficiency shows a median 4% increase during peak optogenetic stimulation, consistent with coordinated vasodilation improving local blood flow distribution." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "synthesis-individual-vessel-measurements-insufficient", "relationType": "cito:confirms" }, { "xref": "responder-threshold-2sd-untested", "relationType": "claimrel:requires" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:requires" } ], "metadata": { "uuid": "ad12a413-e736-4cb5-908f-e331c3a21478", "concepts": [ "capillary network", "efficiency", "optogenetics", "neurovascular coupling" ] } }, { "type": "Claim", "identifier": "constrictions-deeper-than-dilations", "role": "empirical", "children": [ { "type": "Text", "value": "Constricting capillaries are located on average 37±179 µm deeper in cortex than dilating capillaries following 458 nm optogenetic stimulation at 4.3 mW/mm² (p<1e-4), with dilators tending toward the cortical surface across all stimulation conditions." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:disagreesWith" }, { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "responder-threshold-2sd-untested", "relationType": "claimrel:requires" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:supports" } ], "metadata": { "uuid": "a3ca491f-f3b6-456f-b69c-6f9449183c4a", "concepts": [ "cortical depth", "capillary dilation", "capillary constriction", "optogenetics", "neurovascular coupling" ] } }, { "type": "Claim", "identifier": "dilations-nearer-neurons-than-constrictions", "role": "empirical", "children": [ { "type": "Text", "value": "Following ChR2 optogenetic activation at 458 nm, capillary dilations occur on average 16.1±14.3 µm from the nearest labeled pyramidal neuron while constrictions occur 21.9±14.6 µm away (4.3 mW/mm²; p<1e-4), and this spatial segregation is absent under control 552 nm illumination." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "constrictions-deeper-than-dilations", "relationType": "cito:disagreesWith" }, { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "synthesis-individual-vessel-measurements-insufficient", "relationType": "cito:confirms" }, { "xref": "responder-threshold-2sd-untested", "relationType": "claimrel:requires" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "cito:supports" } ], "metadata": { "uuid": "c1381a0c-47be-4721-9022-ddec7d45b3fd", "concepts": [ "neurovascular coupling", "spatial gradient", "capillary dilation", "capillary constriction", "distance to neuron", "ChR2 optogenetics" ] } }, { "type": "Claim", "identifier": "dl-model-scope-single-pipeline", "role": "scope", "children": [ { "type": "Text", "value": "All neurovascular coupling metrics are derived from a single integrated deep learning pipeline (segmentation → registration → graph analysis); the segmentation model requires GPU inference and was trained on data from one imaging preparation (Thy1-ChR2-YFP mice, 6-12 months), so generalization to other preparations is not demonstrated in this paper." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "unetr-outperforms-ilastik-hd95", "relationType": "claimrel:scopes" }, { "xref": "novas3d-outperforms-ilastik", "relationType": "claimrel:scopes" }, { "xref": "registration-doubles-vessel-count", "relationType": "claimrel:scopes" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:scopes" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "claimrel:scopes" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "claimrel:scopes" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "claimrel:scopes" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "claimrel:scopes" }, { "xref": "novas3d-generalizes-qualitatively-ood", "relationType": "claimrel:scopes" } ], "metadata": { "uuid": "6fb7137d-c89c-4651-9271-014bd6823995", "concepts": [ "pipeline scope", "model generalization", "Thy1-ChR2-YFP", "GPU requirement" ] } }, { "type": "Claim", "identifier": "hypothesis-dl-pipeline-enables-network-nvc", "role": "hypothesis", "children": [ { "type": "Text", "value": "A deep-learning segmentation pipeline (UNet/UNETR ensemble) coupled with cross-time-point registration, vertex-wise radius estimation, and graph-theoretic network analysis can produce reliable, automated, network-level measurements of neurovascular coupling in volumetric two-photon fluorescence microscopy of cerebral microcirculation — measurements that prior point-measurement and manual-segmentation approaches could not yield at the scale of hundreds of interconnected vessels per FOV. The performance ceiling of conventional baselines (ilastik, single-time-point segmentation, point-caliber tracking) is set by their inability to handle volumetric SNR, transient RBC plugs, and within-vessel radius heterogeneity — not by an intrinsic limit of the imaging modality." } ], "relations": [ { "xref": "prediction-pipeline-outperforms-baselines", "relationType": "claimrel:entails" }, { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:entails" } ], "metadata": { "uuid": "7c9a1f6b-2b8c-4e1a-9d31-3f0b6c1e8a52", "concepts": [ "deep learning segmentation", "neurovascular coupling", "vascular network", "two-photon fluorescence microscopy", "methodological hypothesis" ], "displayClaim": "A DL segmentation + registration + graph-analysis pipeline can deliver automated network-level neurovascular-coupling measurements across hundreds of vessels per FOV, where prior baselines (ilastik, point-caliber, single-time-point) fall short.", "shortClaim": "A DL pipeline automates network-level neurovascular coupling beyond prior baselines." } }, { "type": "Claim", "identifier": "hypothesis-network-level-nvc-coordination", "role": "hypothesis", "children": [ { "type": "Text", "value": "Optogenetic activation of cortical pyramidal neurons elicits coordinated, network-level vascular responses that cannot be predicted from individual-vessel measurements alone: dilations and constrictions are spatially organised relative to active neurons, capillary responses correlate with their network neighbours' responses, and overall capillary network efficiency is modulated by stimulation. Prior point-measurement studies that interrogate vessels in isolation systematically miss this coordination because they cannot resolve within-vessel radius heterogeneity, between-vessel network topology, or stimulation-evoked changes in graph-level metrics across hundreds of interconnected vessels." } ], "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:entails" }, { "xref": "synthesis-individual-vessel-measurements-insufficient", "relationType": "claimrel:entails" } ], "metadata": { "uuid": "9a1c4f3e-5d2b-4f8a-8c71-1e6f2b9d0a77", "concepts": [ "neurovascular coupling", "network coordination", "optogenetic stimulation", "vascular network topology", "functional hyperemia" ], "displayClaim": "Optogenetic neuronal activation drives coordinated, network-level vascular responses (spatially organised dilations/constrictions, neighbour-correlated capillary responses, stimulation-modulated efficiency) — point measurements of individual vessels cannot resolve this coordination.", "shortClaim": "Optogenetic activation drives coordinated network-level vascular responses across vessels." } }, { "type": "Claim", "identifier": "network-assortativity-increases-stimulation", "role": "empirical", "children": [ { "type": "Text", "value": "Vascular network assortativity increases by 152 ± 65% at 4.3 mW/mm² optogenetic stimulation relative to control, indicating that high-degree vessels preferentially couple with high-degree vessels during neurovascular responses." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "synthesis-individual-vessel-measurements-insufficient", "relationType": "cito:confirms" }, { "xref": "responder-threshold-2sd-untested", "relationType": "claimrel:requires" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:requires" } ], "metadata": { "uuid": "e93b7cce-2578-4075-8e9b-cc9910fc5fd4", "concepts": [ "network assortativity", "optogenetics", "neurovascular coupling", "graph theory" ] } }, { "type": "Claim", "identifier": "novas3d-generalizes-qualitatively-ood", "role": "empirical", "children": [ { "type": "Text", "value": "The NOVAS3D segmentation model produces qualitatively reasonable vessel segmentations on out-of-distribution data including a different mouse strain (C57BL/6), a different species (Fischer rat), and a different microscope modality (light-sheet fluorescence microscopy, Miltenyi UltraMicroscope Blaze), without retraining." } ], "relations": [ { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:confirms" }, { "xref": "dl-model-scope-single-pipeline", "relationType": "cito:extends" } ], "metadata": { "uuid": "afb63ebd-605d-4e1b-a412-9ad7eea3bc58", "concepts": [ "model generalization", "out-of-distribution", "transfer learning", "vessel segmentation", "light-sheet microscopy" ] } }, { "type": "Claim", "identifier": "novas3d-outperforms-ilastik", "role": "control", "children": [ { "type": "Text", "value": "The NOVAS3D deep learning segmentation pipeline (UNet/UNETR) achieves significantly higher Dice scores and precision/recall for volumetric vessel segmentation than the baseline ilastik classifier on the deposited two-photon microscopy test dataset." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:usesMethodIn" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:usesMethodIn" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "cito:usesMethodIn" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "cito:usesMethodIn" }, { "xref": "prediction-pipeline-outperforms-baselines", "relationType": "claimrel:tests" }, { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:confirms" }, { "xref": "unetr-outperforms-ilastik-hd95", "relationType": "cito:extends" } ], "metadata": { "uuid": "73b97665-ce32-4361-81d0-40fb149cea2b", "concepts": [ "NOVAS3D", "UNet", "UNETR", "Dice score", "vessel segmentation", "ilastik baseline" ] } }, { "type": "Claim", "identifier": "novas3d-single-preparation-scope", "role": "scope", "children": [ { "type": "Text", "value": "The NOVAS3D segmentation model was trained and quantitatively evaluated exclusively on data from Thy1-ChR2-YFP mice (line 18, 6–12 months, Texas Red vascular labeling, cranial window, isoflurane anesthesia); quantitative performance metrics are not reported for any other preparation, strain, species, or imaging modality." } ], "epistemicStrength": "strong", "relations": [ { "xref": "unetr-outperforms-ilastik-hd95", "relationType": "claimrel:scopes" }, { "xref": "novas3d-outperforms-ilastik", "relationType": "claimrel:scopes" }, { "xref": "registration-doubles-vessel-count", "relationType": "claimrel:scopes" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:scopes" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "claimrel:scopes" }, { "xref": "baseline-intra-vessel-radius-varies-24pct", "relationType": "claimrel:scopes" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "claimrel:scopes" }, { "xref": "constrictions-deeper-than-dilations", "relationType": "claimrel:scopes" }, { "xref": "blue-light-dilations-exceed-green-control", "relationType": "claimrel:scopes" }, { "xref": "artery-dilates-venule-unchanged-at-low-power", "relationType": "claimrel:scopes" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "claimrel:scopes" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "claimrel:scopes" }, { "xref": "novas3d-generalizes-qualitatively-ood", "relationType": "claimrel:scopes" }, { "xref": "dl-model-scope-single-pipeline", "relationType": "cito:extends" } ], "metadata": { "uuid": "e52585f7-4764-4444-9919-1c9a4c6cec38", "concepts": [ "model scope", "training distribution", "preparation specificity", "generalization" ] } }, { "type": "Claim", "identifier": "prediction-pipeline-outperforms-baselines", "role": "prediction", "children": [ { "type": "Text", "value": "If the DL segmentation + registration + radius-estimation stack genuinely improves on conventional baselines, then (a) the UNet/UNETR ensemble should significantly outperform ilastik on volumetric segmentation metrics (HD95, Dice, precision), (b) cross-time-point registration combined with mask union should recover substantially more vessel segments per FOV than any single time-point segmentation, and (c) the boundary-detection radius estimator should recover known simulated radii with high R² and remain stable to physiologically realistic noise levels." } ], "panel": [ "prediction" ], "relations": [ { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:citesAsSourceDocument" } ], "metadata": { "uuid": "4e5d8b2c-7a91-4c3f-b1e0-2d3f4a5b6c7e", "concepts": [ "benchmark prediction", "pipeline validation", "segmentation metrics", "registration", "radius estimation" ], "displayClaim": "The DL pipeline should beat ilastik on segmentation metrics, registration + mask union should roughly double recovered vessel counts per FOV, and radius estimation should match simulated ground truth at high R² and tolerate realistic noise." } }, { "type": "Claim", "identifier": "prediction-pipeline-reveals-network-coordination", "role": "prediction", "children": [ { "type": "Text", "value": "If the pipeline can resolve network-level neurovascular coupling and if optogenetic activation drives coordinated vascular responses, then in Thy1-ChR2-YFP mice under 458 nm photostimulation we should observe (a) a spatial gradient of dilations vs constrictions relative to labelled neurons, (b) a network-wide rise in assortativity (high-degree vessels coupling with high-degree vessels), (c) a measurable change in capillary network efficiency, and (d) absence of these patterns under 552 nm control illumination and in wild-type mice — none of which would be visible from point-caliber measurements alone." } ], "panel": [ "prediction" ], "relations": [ { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:citesAsSourceDocument" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:citesAsSourceDocument" } ], "metadata": { "uuid": "6f7a8b9c-3e4d-5a6b-7c8d-9e0f1a2b3c4d", "concepts": [ "network coordination prediction", "optogenetic activation", "assortativity", "efficiency", "control conditions" ], "displayClaim": "Under ChR2 photostimulation we should see a dilation-vs-constriction spatial gradient relative to neurons, network-wide assortativity rise, capillary efficiency change — and none of these in green-light or wild-type controls. Point measurements should miss all of them." } }, { "type": "Claim", "identifier": "radius-estimation-r2-0p68", "role": "methodological", "children": [ { "type": "Text", "value": "The pipeline's boundary-detection radius estimation achieves R²=0.68 against ground-truth simulated radii across >100,000 simulations spanning 0.5×–2× dilation/constriction, and remains stable under Gaussian noise up to a standard deviation of 200 signal units (image range 0–1023 SU)." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:usesMethodIn" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "cito:usesMethodIn" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "cito:usesMethodIn" }, { "xref": "prediction-pipeline-outperforms-baselines", "relationType": "claimrel:tests" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:confirms" }, { "xref": "baseline-intra-vessel-radius-varies-24pct", "relationType": "cito:confirms" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:confirms" }, { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:confirms" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "cito:supports" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:supports" } ], "metadata": { "uuid": "9639b9cb-4a39-4046-ba35-db6f8f88dfbb", "concepts": [ "radius estimation", "boundary detection", "validation", "noise robustness", "R-squared" ] } }, { "type": "Claim", "identifier": "registration-doubles-vessel-count", "role": "methodological", "children": [ { "type": "Text", "value": "Rigid registration across time points and union of segmentation masks increases the number of identified vessel segments per FOV from 241±174 (single time point) to 412±281 (union across all time points; 507×507×250 µm, n=107 FOVs), while reducing mean squared error between acquisitions from 1306±747 to 0.008±0.003 signal units." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "network-assortativity-increases-stimulation", "relationType": "cito:usesMethodIn" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "cito:usesMethodIn" }, { "xref": "prediction-pipeline-outperforms-baselines", "relationType": "claimrel:tests" }, { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:confirms" }, { "xref": "dl-model-scope-single-pipeline", "relationType": "cito:supports" } ], "metadata": { "uuid": "2d67695f-3ee3-45e9-b4bc-ff547691c4bc", "concepts": [ "image registration", "vessel extraction", "segmentation union", "transient RBC plug", "capillary gap" ] } }, { "type": "Claim", "identifier": "responder-threshold-2sd-untested", "role": "scope", "children": [ { "type": "Text", "value": "Vessels are classified as responders if their radius change exceeds twice the baseline standard deviation (2×σ); this threshold is not sensitivity-tested in the main analysis, though an alternative 10% threshold is shown in Appendix 1—figure 14 with qualitatively similar results." } ], "relations": [ { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "claimrel:scopes" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "claimrel:scopes" }, { "xref": "constrictions-deeper-than-dilations", "relationType": "claimrel:scopes" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "claimrel:scopes" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "claimrel:scopes" }, { "xref": "blue-light-dilations-exceed-green-control", "relationType": "claimrel:scopes" } ], "metadata": { "uuid": "ce04789d-9922-49de-b9e2-6cc87302b3f4", "concepts": [ "responder threshold", "sensitivity analysis", "methodological assumption", "vessel classification" ] } }, { "type": "Claim", "identifier": "scope-pipeline-and-application-paper", "role": "scope", "children": [ { "type": "Text", "value": "All findings are scoped to a method-paper-with-application envelope: a single deep-learning pipeline (UNet/UNETR ensemble + ANTsPy registration + boundary-detection radius + NetworkX graph analysis) trained and quantitatively benchmarked on Thy1-ChR2-EYFP mice (line 18, 6–12 months, Texas Red vascular labelling, cranial window over S1FL, isoflurane for structural scan, alpha-chloralose for functional scan); the in vivo demonstration uses the same single preparation (n=17 mice for biology, n=4 C57BL/6J wild-type mice for negative control). Quantitative segmentation metrics are reported for nine test images from six held-out mice; out-of-distribution generalisation (C57BL/6, Fischer rat, light-sheet) is shown only qualitatively. The paper does not test the pipeline on awake-behaving preparations, non-mammalian species, disease models, or alternative vascular labels; it does not test alternative responder thresholds beyond a single appendix supplement." } ], "panel": [ "scope" ], "relations": [ { "xref": "unetr-outperforms-ilastik-hd95", "relationType": "claimrel:scopes" }, { "xref": "novas3d-outperforms-ilastik", "relationType": "claimrel:scopes" }, { "xref": "registration-doubles-vessel-count", "relationType": "claimrel:scopes" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:scopes" }, { "xref": "vessel-radius-heterogeneity-stimulation", "relationType": "claimrel:scopes" }, { "xref": "baseline-intra-vessel-radius-varies-24pct", "relationType": "claimrel:scopes" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "claimrel:scopes" }, { "xref": "constrictions-deeper-than-dilations", "relationType": "claimrel:scopes" }, { "xref": "blue-light-dilations-exceed-green-control", "relationType": "claimrel:scopes" }, { "xref": "artery-dilates-venule-unchanged-at-low-power", "relationType": "claimrel:scopes" }, { "xref": "network-assortativity-increases-stimulation", "relationType": "claimrel:scopes" }, { "xref": "capillary-efficiency-increases-4pct", "relationType": "claimrel:scopes" }, { "xref": "wt-controls-no-blue-green-difference", "relationType": "claimrel:scopes" }, { "xref": "novas3d-generalizes-qualitatively-ood", "relationType": "claimrel:scopes" } ], "metadata": { "uuid": "1a2b3c4d-5e6f-7a8b-9c0d-1e2f3a4b5c6d", "concepts": [ "method-paper scope", "preparation envelope", "training distribution", "benchmark distribution", "OOD evaluation" ], "displayClaim": "Pipeline-and-application scope: one DL stack trained on Thy1-ChR2-EYFP cranial-window 2PFM, benchmarked on six held-out mice, applied in vivo to n=17 Thy1-ChR2 + n=4 WT mice. OOD shown qualitatively only; no awake behaviour, no disease models, no alternative thresholds." } }, { "type": "Claim", "identifier": "synthesis-individual-vessel-measurements-insufficient", "role": "synthesis", "children": [ { "type": "Text", "value": "Interrogating individual vessels in isolation is insufficient to predict how blood flow is modulated across the cerebral microvascular network. The paper's combined evidence — within-vessel radius heterogeneity (24 ± 28% along vessel length at baseline), the spatial segregation of dilations and constrictions relative to active neurons, the network-wide rise in assortativity during optogenetic stimulation, and the modulation of capillary network efficiency — together argue that point-caliber and single-vessel approaches systematically miss the coordinated, graph-level structure of functional hyperemia." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:citesAsSourceDocument" } ], "metadata": { "uuid": "8b9c0d1e-2f3a-4b5c-6d7e-8f9a0b1c2d3e", "concepts": [ "synthesis", "network-level neurovascular coupling", "methodological argument", "point-measurement insufficiency" ], "displayClaim": "Single-vessel measurements cannot predict network-level blood flow modulation: within-vessel heterogeneity, neuron-relative spatial gradients, assortativity rise, and efficiency changes together require a network-level treatment.", "shortClaim": "Single-vessel measurements cannot predict network-level blood flow modulation." } }, { "type": "Claim", "identifier": "unetr-outperforms-ilastik-hd95", "role": "control", "children": [ { "type": "Text", "value": "UNETR ensemble segmentation shows significantly better HD95 surface distance than ilastik for both vessel and neuron channels (p<0.05 Wilcoxon signed-rank), while ilastik over-segments vessels with high recall (0.89±0.19) but low precision (0.37±0.33), evaluated on nine test images (507×507×250 µm) from six held-out mice." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "novas3d-outperforms-ilastik", "relationType": "cito:disagreesWith" }, { "xref": "prediction-pipeline-outperforms-baselines", "relationType": "claimrel:tests" }, { "xref": "hypothesis-dl-pipeline-enables-network-nvc", "relationType": "cito:confirms" }, { "xref": "novas3d-single-preparation-scope", "relationType": "claimrel:requires" }, { "xref": "novas3d-outperforms-ilastik", "relationType": "cito:extends" } ], "metadata": { "uuid": "db0a9448-8fe8-4336-b220-0a71aa19d870", "concepts": [ "UNETR", "ilastik", "HD95", "vessel segmentation", "Dice score", "precision", "recall" ] } }, { "type": "Claim", "identifier": "vessel-radius-heterogeneity-stimulation", "role": "empirical", "children": [ { "type": "Text", "value": "Vessel radius adjustments during optogenetic stimulation show 24 ± 28% variation relative to resting diameter, with dilations averaging 16.1 ± 14.3 µm and constrictions averaging 21.9 ± 14.6 µm relative to nearby neurons." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "radius-estimation-r2-0p68", "relationType": "claimrel:requires" }, { "xref": "responder-threshold-2sd-untested", "relationType": "claimrel:requires" } ], "metadata": { "uuid": "de368a2f-dfdf-4cb5-9c5f-24e3e0cf3aae", "concepts": [ "vessel radius", "dilation", "constriction", "optogenetics", "heterogeneity" ] } }, { "type": "Claim", "identifier": "wt-controls-no-blue-green-difference", "role": "control", "children": [ { "type": "Text", "value": "Wild-type C57BL/6J mice (n=4) show no statistically distinguishable capillary radius distributions following blue (458 nm) versus green (552 nm) photostimulation, confirming that vascular responses in Thy1-ChR2 mice are ChR2-specific and not attributable to photothermal or non-specific light effects." } ], "epistemicStrength": "moderate", "relations": [ { "xref": "prediction-pipeline-reveals-network-coordination", "relationType": "claimrel:tests" }, { "xref": "blue-light-dilations-exceed-green-control", "relationType": "cito:confirms" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:confirms" }, { "xref": "constrictions-deeper-than-dilations", "relationType": "cito:confirms" }, { "xref": "hypothesis-network-level-nvc-coordination", "relationType": "cito:confirms" }, { "xref": "blue-light-dilations-exceed-green-control", "relationType": "cito:supports" }, { "xref": "dilations-nearer-neurons-than-constrictions", "relationType": "cito:supports" } ], "metadata": { "uuid": "e05b0913-66d9-4ac2-be33-40385aaf6181", "concepts": [ "wild-type control", "ChR2 specificity", "photothermal artifact", "C57BL/6J", "negative control" ] } } ] } ], "title": [ { "type": "Text", "value": "A deep learning pipeline for mapping in situ network-level neurovascular coupling in multi-photon fluorescence microscopy" } ] }