Ultrafast Doppler Vascular Imaging for Intraoperative Physiological Monitoring of Human Spinal Cord

Summary

This study presents the application of ultrafast Doppler vascular imaging for physiological monitoring during human spinal cord surgery.

Abstract

Monitoring physiological parameters is essential for assessing the patient's condition during intraoperative scenarios. Hemodynamic information, such as blood pressure and heart rate, can be derived from ultrasound Doppler signals, reflecting vascular function and indicating pathological risks in real-time. Compared to oscillometry, magnetic resonance imaging (MRI), conventional Doppler, and piezoelectric methods, ultrafast Doppler provides superior temporal resolution and high sensitivity for small vessels. In this study, an intraoperative application of ultrafast Doppler imaging in the spinal cord of a patient with Chiari malformation is described. Post-processing of the data generated power Doppler images of the spinal cord vasculature and physiological parameters, including detailed mapping of the resistivity index (RI) and pulsatility index (PI). Both RI and PI showed significant reductions after surgical intervention (RI mean: 0.47 → 0.32; PI mean: 0.63 → 0.39; p < 0.001), with consistent trends observed in the medians, indicating improved spinal cord hemodynamics. Evaluation of changes in PI/RI parameters using this method may offer a broader understanding of disease progression and treatment efficacy, potentially guiding surgical strategies and therapeutic approaches.

Introduction

Blood pressure dynamics, as an important intraoperative physiological indicator, provides valuable guidance for surgical strategy and ensures operational safety¹. Pulse waves, generated by the periodic contraction and relaxation of the heart, propagate through the arterial vascular system from the aorta as pressure waves². These waves can be regarded as a direct manifestation of cardiovascular function, offering critical information for understanding blood pressure dynamics and overall hemodynamics³. In clinical practice, pulse waves are extensively utilized for continuous assessment of hemodynamic status⁴.

As a relay station and reflex center within the central nervous system, the spinal cord plays a pivotal role in transmitting motor and sensory information. The stability of its blood supply is crucial for maintaining and restoring neurological function⁶. The spinal cord pulsatile blood flow signal is closely associated with the pulse wave, containing information about vascular elasticity and hemodynamics, thereby reflecting the physiological and pathological changes of the spinal cord. Monitoring these signals faces considerable challenges due to the spinal cord’s small size, intricate vascular architecture⁷, and complex blood flow patterns⁸,  which complicate effective assessment of spinal cord pulse waves.

Several techniques for assessing pulse waves have been proposed. The oscillometric analysis method is easy to operate and provides rapid measurements. However, the accuracy of the results may be influenced by factors such as artifacts, patient posture, and the appropriate size and placement of the cuff^9,10. Some piezoelectric biosignal sensors, such as piezoelectric pulse sensors, can efficiently detect pulse waves, yet their accuracy may be compromised by mechanical vibrations or temperature fluctuations; meanwhile, optical sensors such as photoplethysmography (PPG) are susceptible to environmental light and motion artifacts¹¹. Furthermore, these techniques can only capture localized pulse wave signals and lack the ability to provide imaging information of the observed structure simultaneously¹².

As a powerful imaging modality, magnetic resonance imaging (MRI) offers relatively high spatial resolution measurement of the spinal cord's arterial walls, but it is costly and has limited accessibility¹³. Computed tomography (CT) imaging provides faster scan times and broader availability but emits ionizing radiation, which restricts its use for repeated assessments¹⁴. Conventional Doppler ultrasound enables real-time assessment of blood flow dynamics through two complementary imaging modes: color Doppler and pulse wave Doppler (PWD). Color Doppler provides a qualitative overview of flow velocities across the field of view by scanning the medium line-by-line, but suffers from low frame rates and poor sensitivity to slow or deep flows. PWD offers quantitative temporal information about blood flow waveforms, but is limited to a single, user-defined region of interest. As a result, it is applicable for strong hemodynamics in arteries and veins and quantifies instantaneous pulse waveforms within a small field of view¹⁵. Contrast-enhanced ultrasound improves the visualization of blood flow through the use of microbubble agents; however, the requirement for contrast injection raises safety concerns in intraoperative settings, rendering it less suitable for real-time assessment of vascular pulsatility¹⁶. Photoacoustic and optical imaging techniques provide high contrast and spatial resolution for vascular imaging, particularly in superficial tissues. However, their strong dependence on optical access significantly limits their applicability to deep-seated structures such as the spinal cord¹⁷.

In recent years, ultrafast ultrasound has advanced rapidly with great potential in vascular imaging¹⁸. Unlike traditional focused beam scanning, ultrafast Doppler uses multi-angle compounded plane wave techniques, which enhance ultrasound imaging sensitivity by about 50-fold and highly improve the detection of small vessels^15,19,20. A complete Doppler spectrum can be obtained for each pixel in the image, facilitating the acquisition of comprehensive information on blood flow velocity and direction²¹. This can be used to estimate vascular indices such as resistivity index (RI) and pulsatility index (PI) throughout the imaging area. Notably, in 2014, Demene et al.²² applied ultrafast Doppler technology to neonatal cerebral blood flow imaging, presenting a detailed distribution map of cerebral vascular RI in neonates. This advancement provides a new tool for understanding the mechanisms of cerebral blood flow autoregulation and the pathogenesis of related diseases in preterm and term infants. In 2020, Bourquin et al.^23,24 proposed dynamic ultrasound localization microscopy (DULM) based on ultrafast ultrasound technology to achieve in vivo measurements of pulsatile microcirculation in the rodent brain and extended this technique to three-dimensional imaging, offering a new dimension for quantitative analysis.

Towards vascular imaging of the spinal cord, in 2021, Zang et al.²⁵ achieved spinal cord microvascular imaging without contrast agents using ultrafast Doppler, with an imaging resolution comparable to the transmitted wavelength. Sui et al.²⁶. employed a random sampling method based on robust principal component analysis (RPCA) to achieve fast, high signal-to-noise ratio ultrafast Doppler imaging of microcirculation in the brain and spinal cord. Pezet et al.²⁷ observed vascular reconstruction following chronic spinal cord injury. In 2022, Yu et al.²⁸ achieved ultrafast super-resolution ultrasound localization microscopy of spinal cord microcirculation in rats based on RPCA, with imaging resolution reaching 13-16 µm, significantly smaller than the 100 µm wavelength. Further studies involved continuous observation and quantitative analysis of microcirculation in the spinal cord penumbra following spinal cord injury²⁹. In 2023, Yan et al.³⁰ utilized ultrafast ultrasound vector Doppler to perform vectorized imaging of small-vessel blood flow in human spinal cord tumors and assessed sequence-related safety in detail based on experimental sequence parameters. In 2024, Agyeman et al.³¹ conducted the first functional ultrasound imaging of the spinal cord in humans, which demonstrated the integration of spinal cord functional responses to electrical stimulation. Khaing et al.³² validated the correlation between perfusion imaging indicators and injury severity in both rats and humans following acute traumatic spinal cord injury. However, the detection and analysis of hemodynamic indices in spinal cord pulse waves still require further investigation.

Chiari malformation represents a significant and prevalent disorder impacting the spinal cord. It induces dynamic alterations in cerebrospinal fluid and spinal cord compression, disrupting spinal cord blood flow and vascular regulation, which may result in abnormal venous drainage³³, increased vascular resistance³⁴, and potential spinal cord ischemia³⁵. Observation of spinal cord pulse waves facilitates a deeper understanding of spinal cord hemodynamic changes, offering valuable insights for diagnosing and treating Chiari malformation and other spinal cord disorders³⁶.

This study presents a comprehensive protocol for utilizing ultrafast Doppler vascular imaging to intraoperatively monitor spinal cord physiological parameters in humans, with a specific focus on its application in a patient with Chiari malformation. This method enables real-time acquisition of high-frame-rate ultrafast Doppler data during exposure of the spinal cord, followed by post-processing to generate pixel-wise maps of RI and PI.

Protocol

This study involved human subjects, and all procedures were conducted in accordance with protocols approved by the Human Research Ethics Committee, Huashan Hospital affiliated with Fudan University (2021-065). Informed consent was obtained from all participants. The implementation was solely intended for preliminary research and validation purposes, not for clinical diagnosis. Relevant FDA-recommended safety indices were assessed to ensure compliance with established safety standards. Patients included in this study were diagnosed with Chiari malformation. The inclusion, exclusion, and withdrawal criteria for participants are provided in Supplementary File 1. The reagents and equipment used are listed in the Table of Materials.

1. Instrument preparation

1. Prepare the ultrasound imaging system.
   1. Connect the 128-channel linear array to the programmable ultrasound system.
      NOTE: Protect the ultrasound probe using a custom-designed housing, modeled with 3D software and fabricated by 3D printing. The probe used in this study was a miniaturized customized version rather than a commercial version. The ultrasound system platform employed in this study has previously been applied in several studies for the investigation of human vascular imaging^37,38,39,40.
   2. Set the center frequency of the transducer to 15.625 MHz, corresponding to 100 µm spatial resolution.
   3. Ensure the pitch between adjacent elements is configured as 0.1 mm.
2. Calibrate the device, test the code, and ensure that the system is operating correctly.
3. Set up a platform for stable imaging.
   1. Position the robotic arm beside the surgical bed and lock it securely in place for subsequent probe fixation.
   2. Ensure that the probe remains fixed throughout the imaging session.
      NOTE: This setup minimizes shaking and positional shifts caused by manual operation, improving the stability and consistency of probe positioning during imaging.

2. Ultrasound sequence preparation

1. Set an appropriate imaging depth according to the B-mode ultrasound imaging.
2. Set the compounded frame rate to 1000 Hz.
3. Set the acquisition time to 0.4 s.
4. Transmit 11 tilted plane waves uniformly spaced between -10° and +10°, with a pulse repetition frequency of 11,000 Hz.
   NOTE: Tilted plane waves are emitted at various angles relative to the probe's lateral axis. In this study, 11 plane waves were evenly distributed from -10° to +10°.

3. Probe positioning

1. Cover the ultrasound probe with a sterile protective bag for medical instruments (Figure 1A).
   1. Apply a generous amount of ultrasound gel inside the sterile bag.
   2. Insert the probe into the bag.
   3. Secure the probe inside the bag with a rubber band.
      NOTE: Acoustic coupling was achieved by applying sterile ultrasound gel to the inner surface of the sterile bag, which enclosed the probe. The bag itself was placed directly in contact with human tissue, allowing effective ultrasound transmission without direct contact between the coupling medium and the tissue.
2. Secure the probe with the robotic arm (Figure 1B).
   1. Position the probe precisely on the exposed surface of the spinal cord following laminectomy.
   2. Stabilize the probe using the robotic arm to maintain consistent positioning during imaging.
      NOTE: The pneumatic robotic arm was used to stabilize the ultrasound probe. Although it lacks automatic scanning capabilities, the arm enables precise manual positioning and stability throughout the data acquisition process. The arm, equipped in the operating room, was manually controlled and fixed by the primary surgeon. At first, all major joints were secured, followed by fine adjustment of the terminal joint that holds the probe under B-mode ultrasound guidance.

4. Data acquisition

1. Set an appropriate imaging depth according to the B-mode ultrasound imaging.
2. Use real-time B-mode ultrasound imaging to identify the appropriate imaging plane.
3. Locate the herniated cerebellar tonsils and spinal cord.
   NOTE: Confirm proper alignment when both the spinal cord and the cerebellar tonsils are clearly visualized within the same imaging plane, with distinguishable boundaries and appropriate anatomical orientation. In the B‑mode imaging plane, ensure that the cerebellar tonsils appear superior to the spinal cord, and that the dorsal margin of the cord is clearly defined with left–right symmetry. After confirming the correct imaging plane, acquire a single ultrafast Doppler image to verify the presence and stability of blood flow signals.
4. Terminate the B-mode sequence and initiate the ultrafast sequence for data acquisition.
   NOTE: Perform the procedure under normal room temperature conditions. The imaging duration is short, and the probe did not exhibit any noticeable temperature increase during acquisition. This method provides a clear visualization of spinal cord structures during surgery, allowing for precise probe adjustments to achieve optimal vascular imaging, and enables comparison with preoperative MRI images. In the case presented in this study, the original imaging depth range was 0.5 to 9.5 mm. The actual imaging depth was held constant across all panels.

5. Analysis of spinal cord pulsatile blood flow signal before therapy (Figure 2)

1. Generate power Doppler images of spinal cord vasculature. (Figure 3)
   NOTE: Probe stability was evaluated through power Doppler imaging analysis. The consistent visualization of spatially distinct vessels devoid of smearing artifacts demonstrates the efficacy of the robotic arm in preserving a stable acoustic window during acquisition.
   1. Perform beamforming on the acquired radiofrequency data.
   2. Apply the singular value decomposition (SVD)-based spatiotemporal filtering method to separate tissue signals from blood flow signals and remove high-frequency noise. Set appropriate thresholds to extract the desired blood flow components.
      ​NOTE: The ultrasound image signal comprises components of soft tissue, blood flow, and random noise. SVD-based spatiotemporal filtering separates these components based on their spatiotemporal characteristics⁴¹. This method decomposes the ultrasound image data into singular vectors and their corresponding singular values. By setting a filtering threshold of the singular values, large singular values are associated with eigenvectors representing tissue components with strong spatiotemporal correlation, medium singular values correspond to dynamic blood flow signals, and small singular values mainly represent noise. In this study, the SVD thresholds were set to retain singular components 80-380 prior to electrocautery treatment and 50-380 after surgical intervention. This threshold was empirically determined based on prior experience (Supplementary Figure 7).
   3. Demodulate the filtered blood flow signal into in-phase/quadrature (IQ) complex signals . Compute the mean signal intensity at each pixel to obtain the power Doppler image PW (x,z) (Equation 1)¹⁵. N_t is the number of frames acquired.
      (1)
      NOTE: Eligible Doppler data are characterized by clear vascular depiction and a high signal-to-noise ratio. If vascular visualization is suboptimal, revisit step 5.1.2 and adjust the SVD threshold range accordingly.
2. Perform time-frequency analysis of the Doppler signal (Figure 4).
   1. Segment the blood flow signal using a short-time window (Figure 4A).
   2. Apply the Fast Fourier Transform (FFT) to estimate power spectral density at each pixel (Figure 4B).
      ​NOTE: The window length was chosen to sufficiently cover at least one cardiac cycle, ensuring that both systolic and diastolic frequency components could be reliably identified. A Hann window was used to reduce spectral leakage, and adjacent windows were overlapped to enhance temporal resolution. In this study, FFT was applied using a Hann window of length 80 with a 70% overlap.
   3. Slide the window continuously to obtain the time-varying mean Doppler frequency.
3. Calculate RI and PI maps (Figure 4C and Figure 5A-D).
   1. Analyze the average Doppler frequency of blood flow signals over time at specific pixels (Figure 4C).
      NOTE: The validity of frequency analysis can be confirmed by observing periodic fluctuations in the average Doppler frequency that correspond to the cardiac cycle.
   2. Determine blood flow velocity and its variations to calculate RI (Equation 2)¹⁵ and PI (Equation 3)¹⁵. V_PSV is peak systolic velocity, V_EDV is end-diastolic velocity, and V_mean is the mean flow velocity.
      (2)
      (3)
   3. Simplify the RI and PI calculation using Doppler frequency. Calculate RI (Equation 4) and PI (Equation 5) across the entire spinal cord. f_mean(x,z,t) is the average Doppler frequency within a defined time window¹⁵.
      (4)
      (5)
      NOTE: Due to blood flow, the ultrasound signal undergoes a frequency deviation known as Doppler shift, which is related to the blood flow velocity V (Equation 6). f_Doppler represents the Doppler frequency shift, c₀ denotes the speed of sound in human soft tissue, f_pulse is the transmitted frequency of the ultrasound probe, and is the angle between the ultrasound beam and the direction of blood flow¹⁵.
      (6)
   4. The average Doppler frequency represents the mean value of the shift within a specific time window. Since V is proportional to the Doppler shift, it can be approximated by evaluating the average Doppler frequency at each pixel over time.
4. Statistical analysis of RI and PI
   1. Use logical masks to extract valid RI and PI values from regions of interest, and exclude outliers using the interquartile range (IQR) method.
   2. Perform a weighted Student's t-test to evaluate differences between pre- and post-surgical conditions, accounting for the pixel-wise data structure.
      ​NOTE: Statistical analyses of RI and PI were conducted at the pixel level, and group differences were visualized using boxplots with significance levels indicated by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001).

6. Patient management and safety of imaging protocols

1. Monitor vital signs closely. Continuously monitor heart rate, blood pressure, respiratory rate, oxygen saturation, and body temperature during imaging. Ensure the patient's safety and comfort throughout the procedure.
   NOTE: Intraoperative neurophysiological monitoring (such as somatosensory evoked potentials and motor evoked potentials) is also conducted, considering nerve damage may occur during the surgery.
2. Ensure appropriate probe pressure. Use the robotic arm to maintain stable and adequate pressure on the probe. Avoid excessive force that could damage the surface of the spinal cord.
3. Maintain communication with the surgical team. Communicate clearly with the surgical team throughout the imaging process. Respond promptly to any patient discomfort or unexpected situations.

7. Acquisition and analysis of spinal cord pulsatile blood flow signal after therapy

1. Repeat the steps described in step 5 to analyze the spinal cord blood flow signal after treatment.
   NOTE: After electrocautery treatment, the cerebellar tonsils retracted with reduced volume, blunted edges, and restored anatomical relationships with surrounding tissue structures (Figure 6). Data acquisition was performed at two distinct intraoperative time points: prior to electrocautery treatment and after surgical intervention. Currently, the exploration of RI and PI remains at a preliminary stage and has not been implemented for real-time surgical guidance. As a result, shortly after rapid data acquisition, the surgical procedure continued without significant interruption. Then, the collected data were processed offline (approximately 5-10 min,details of the processing steps and timing are provided in Supplementary File 2), ensuring no impact on the surgical workflow. In the future, if RI and PI can be used to assist intraoperative decision-making, they are expected to be integrated without disrupting the surgical timeline.

Results

Ultrafast Doppler offers high sensitivity, enabling high temporal resolution imaging of the spinal cord vasculature network. Compared to color Doppler, power Doppler offers a higher signal-to-noise ratio (SNR)⁴². An appropriate threshold range was used for the SVD-based spatiotemporal filter, enabling the identification of valid blood flow signals and the imaging of blood flow direction. Figure 3E,F presents the power Doppler images of the spinal cord...

Discussion

In this study, ultrafast Doppler imaging was employed intraoperatively to extract the blood flow of the human spinal cord. Spectral analysis and calculation of the RI and PI were subsequently performed on the blood flow signals, providing power Doppler images of spinal cord vasculature and corresponding RI and PI maps of the patient with Chiari malformation. In the context of ultrafast ultrasound, small vessels refer to vessels with diameters on the order of 100 micrometers, while vessels smaller than 100 micrometers wer...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
MATLAB  R2022b	MathWorks	N/A	For post-processing of data
Microscope	ZEISS	PENTERO 8O0 S	For intraoperative microscopic observation
Programmable ultrasound system	Verasonics	Vantage 256	For ultrasound imaging of spinal vessels
Robotic arm	Aesculap	RT060R	For fixing the probe
Single crystal high frequency linear array transducers	Verasonics	L22-14vX	For ultrasound imaging of spinal vessels
Ultrasound gel	Baby Fun	Type M	Eliminate the air gap between the probe and the contact surface