【Nat.Biomed.Eng.】上海交通大学医学院郑元义等|突破139微米分辨率限界！临床转化级超声局域化显微镜揭示脑外伤后血管重构与预后#

a, Schematic of bedside ULM imaging and contrast agent delivery setup in ICU patients with TBI. b, Standardized placement of the ultrasound imaging plane over the craniectomy window using standard anatomical landmarks. c, Schematic illustration of the standardized scanning plane, showing the lateral ventricles, third ventricle and basal ganglia. d, Representative super-resolution trajectory density map. e, Representative cerebral blood flow direction and velocity maps. f, Magnified view of the boxed region in d, showing microvascular structure. g, Cross-sectional profiles of the two vessel pairs marked (i) and (ii) in f. Dots indicate normalized microbubble signal intensity, and dashed lines indicate Gaussian fits. MBs, microbubbles.#

a, B-mode ultrasound image. b, Corresponding cranial CT scan (1-mm slice thickness) image showing the midline, ventricular structures and post-traumatic lesion. c, CTA image with the ultrasound imaging field overlaid. d, Super-resolution ULM density map showing microvascular architecture. e, Zoomed CTA view showing major vessels (blue arrows) and poorly resolved microvessels (green arrows). f, Corresponding zoomed ULM view showing the same major vessels (blue arrows) together with finer vascular hierarchies and branching structures (green arrows). g, CDFI showing sparse signals from major vessels. h, Cerebral blood flow map from CTP. i, ULM-derived cerebral blood flow velocity map showing localized perfusion deficits with higher spatial resolution and flow detail.#

Paired CT and corresponding ULM images are shown from representative patients with TBI. Individual patients may exhibit more than one pattern across different brain regions or at different timepoints. a, Traumatic intracerebral haematoma: the circled hyperdense lesion on CT corresponds to a complete flow void on ULM (n = 5). b, Cerebral contusion or haemorrhage: the circled lesion appears haematoma-like on CT but retains vascular structure and perfusion on ULM (n = 7). c, Post-traumatic cerebral ischaemia: the circled hypodense area on CT without swelling corresponds to regional hypoperfusion on ULM (n = 3). d, Post-traumatic cerebral swelling: the circled hypodense area with diffuse cerebral swelling and MLS on CT corresponds to reduced microvascular flow on ULM (n = 10). e, PTCI: a well-defined hypodensity on CT corresponds to complete vascular dropout on ULM (n = 3). f, PTH: ventricular enlargement on CT is accompanied by global hypoperfusion and distinct flow patterns on ULM (n = 3). Right-hand panels summarize illustrative decision-support interpretations.#

a, Representative ULM vascular and velocity maps from moderate and severe TBI on postoperative days 3 and 14. b, ICP and CPP; n = 8/9 (moderate/severe) on day 3. c–e, Quantitative analyses based on PP/V (c), PP/CMC (d) and vascular morphology (e). c, PP/V across vessel scales. d, PP/CMC, reflecting microvascular-level resistance. e, Branch density, junction density and vascularity of medium vessels. For c–e, n = 8/11 on day 3 and n = 6/12 on day 14. Box plots show interquartile range with median lines; whiskers indicate minimum and maximum values, and dots represent individual patients. Exact P values are shown in the figure. All statistical tests were two-sided. Statistical comparisons were performed using an unpaired two-tailed t-test (b) and two-way ANOVA fitted as a mixed-effects model (c–e) with Benjamini–Hochberg correction for multiple comparisons.#

纵向成像结果揭示了脑损伤后病情严重程度依赖性的微血管重构轨迹。相比于中度损伤患者展现出的良好结构修复，重度损伤患者在早期表现出更高的微血管阻力以及持续的血流灌注缺陷。进一步的数据分析表明，基于微血管平均流速和峰度构建的复合超声指标与临床有创颅内压及脑灌注压表现出强相关性。此外，研究提出的微血管灌注效率系数（MPEC）能够鲁棒地预测患者6个月后的远期神经功能预后，这表明针对微循环的个性化治疗干预有望直接改善患者的康复效果。

a, Anatomical segmentation into cortex, white matter, basal ganglia and peri-impairment area. b, Vascularity in the basal ganglia and cortex. c,d, Branch density (c) and junction density (d) in the peri-impairment region. e, Haemodynamic heterogeneity in the basal ganglia, evaluated using velocity kurtosis (_V_kurtosis) and velocity skewness (_V_skewness). f, Microvascular perfusion quantified by CMC in the white matter. g, Regional microvascular resistance estimated using PP/CMC in the cortex, basal ganglia and peri-impairment regions. Box plots show interquartile range with median lines; whiskers indicate minimum and maximum values, and dots represent individual patients. Outliers are marked as crosses (×). Exact P values are shown in the figure. All statistical tests were two-sided. Statistical comparisons were performed using two-way ANOVA fitted as a mixed-effects model with Benjamini–Hochberg correction for multiple comparisons. Day 3: n = 8/10 (moderate/severe) for all regions except peri-impairment (n = 8/11); day 14: n = 6/11 for all regions except peri-impairment (n = 6/12).#

a, Correlation heat map showing associations between vascular haemodynamic and morphological features and clinical variables across vessel scales and brain regions. b, Scatter plots of ULM-derived parameters (excluding major vessels) exhibiting the strongest positive and negative correlations with ICP and CPP, with linear regression lines. c,d, Performance of a composite metric combining mean velocity and velocity kurtosis in microvessels, demonstrating a strong correlation with CPP (c) and ICP (d). The normalization constants (1.85 and 3.12) represent the population means of the current cohort used for data centring. Pearson correlation coefficients were used for normally distributed data and Spearman correlation coefficients for non-parametric data; all corresponding P values were two-sided. P values for correlations involving invasive parameters (ICP and CPP) with |r| > 0.5 were adjusted for multiple comparisons using the Benjamini–Hochberg method. Exact P values are shown in the figure. n = 16 for basal ganglia and white matter in b–d; n = 17 for medium vessels, peri-impairment and microvessels in b–d.#

a, ICP and CPP in poor and good outcome groups stratified by GOS scores. n = 8/9 (poor/good) on day 3. b, Medium-vessel branch density, junction density and vascularity at day 14. c–e, Region-specific prognostic analyses in the basal ganglia (c), white matter (d) and peri-impairment region (e). f,g, Prognostic performance of the dimensionless composite perfusion efficiency metric (0.5_ρ_ × CMC2/PP) using white matter ULM data at day 3 (f) and global microvascular data at day 14 (g). Box plots show the interquartile range with median lines; whiskers indicate minimum and maximum values and dots represent individual patients. Outliers are marked as crosses (×). Exact P values are shown in the figure. All statistical tests were two-sided. For normally distributed data, statistical comparisons were performed using unpaired Student’s t-tests when variance homogeneity was confirmed and Welch’s t-tests otherwise. For non-normally distributed data, Mann–Whitney tests were used. n = 9/9 (poor/good) on day 3 and n = 10/7 on day 14 in c, d and f; n = 10/9 on day 3 and n = 11/7 on day 14 in b, e and g.#

a, CEUS acquisition using a clinically available system. Multiple 60-s CEUS videos were acquired, and stable data segments were aggregated into a cumulative 180-s dataset for analysis. The plot shows the average number of microbubble localizations per 60-s CEUS video across patients. b, Motion correction by rigid image registration, visualized by vector fields (yellow arrows), with post-registration correlation ≥0.99. c, A representative CEUS frame and corresponding estimated spatially varying PSF (magnified). d, Local blind deconvolution within an 80 × 80 pixel block using the estimated PSF to enhance microbubble localization. e, Structural prior map derived from detected bubble centres to guide subsequent tracking algorithms. f, Bubble tracking algorithm incorporating the structural prior, spatiotemporal motion coherence filtering and globally optimized data association across four time steps. g, Asynchronous parallel processing architecture, with GPU-based distance-tree initialization and CPU-based vectorized tracking tasks (1,000 frames each). MBs, microbubbles; PSF, point spread function; corr, cross-correlation coefficient.#