Photoacoustic monitoring of blood oxygenation during neurosurgical interventions

TL;DR

This study demonstrates that multispectral photoacoustic imaging can monitor blood oxygenation changes in a gyrencephalic brain during neurosurgery, with potential for clinical applications despite some quantification limitations.

Contribution

First in vivo demonstration of multispectral PA imaging for sO2 monitoring in a gyrencephalic brain during neurosurgery, with open-source spectral unmixing algorithms.

Findings

PA imaging can monitor cerebral sO2 in a pig model.

Linear spectral unmixing detects oxygenation changes effectively.

Quantification of sO2 has depth-related limitations.

Abstract

Multispectral photoacoustic (PA) imaging is a prime modality to monitor hemodynamics and changes in blood oxygenation (sO2). Although sO2 changes can be an indicator of brain activity both in normal and in pathological conditions, PA imaging of the brain has mainly focused on small animal models with lissencephalic brains. Therefore, the purpose of this work was to investigate the usefulness of multispectral PA imaging in assessing sO2 in a gyrencephalic brain. To this end, we continuously imaged a porcine brain as part of an open neurosurgical intervention with a handheld PA and ultrasonic (US) imaging system in vivo. Throughout the experiment, we varied respiratory oxygen and continuously measured arterial blood gases. The arterial blood oxygenation (SaO2) values derived by the blood gas analyzer were used as a reference to compare the performance of linear spectral unmixing algorithms in this scenario. According to our experiment, PA imaging can be used to monitor sO2 in the porcine cerebral cortex. While linear spectral unmixing algorithms are well-suited for detecting changes in oxygenation, there are limits with respect to the accurate quantification of sO2, especially in depth. Overall, we conclude that multispectral PA imaging can potentially be a valuable tool for change detection of sO2 in the cerebral cortex of a gyrencephalic brain. The spectral unmixing algorithms investigated in this work will be made publicly available as part of the open-source software platform Medical Imaging Interaction Toolkit (MITK).

Tables

Table 1. Table 1: Comparison of the arterial blood gas ( ABG ), pulse oximetry ( PuOx ) and five spectral unmixing ( SU ) estimations – in region of interest ( ROI ) 1 & 2. Values are averaged over one minute beginning at the time after start of the recording. The respiratory oxygen ( rO 2 ) value was adjusted at least five minutes before the recording (see Figure 2). SVD: singular value decomposition, NNLS: non-negative least squares, WLS: weighted least squares. –: Missing values failed to record due to technical issues.

time	rO2	SaO2 [%]			QR/SVD sO2 [%]		LU sO2 [%]		NNLS sO2 [%]		WLS sO2 [%]
[min]	[%]	ABG	PuOx		ROI 1	ROI 2	ROI 1	ROI 2	ROI 1	ROI 2	ROI 1	ROI 2
+0	35	100	99		85 $\pm$ 2	99 $\pm$ 2	85 $\pm$ 2	98 $\pm$ 2	85 $\pm$ 2	98 $\pm$ 2	85 $\pm$ 2	98 $\pm$ 2
+8	21	93	88-92		72 $\pm$ 3	98 $\pm$ 2	72 $\pm$ 4	98 $\pm$ 2	72 $\pm$ 3	98 $\pm$ 1	72 $\pm$ 3	98 $\pm$ 2
+24	0	26	40		27 $\pm$ 2	84 $\pm$ 3	26 $\pm$ 2	83 $\pm$ 4	27 $\pm$ 2	84 $\pm$ 3	27 $\pm$ 2	84 $\pm$ 3
+34	21	86	–		69 $\pm$ 2	91 $\pm$ 3	69 $\pm$ 2	91 $\pm$ 3	69 $\pm$ 2	91 $\pm$ 3	69 $\pm$ 2	91 $\pm$ 3
+45	100	–	100		95 $\pm$ 2	91 $\pm$ 2	95 $\pm$ 2	90 $\pm$ 2	95 $\pm$ 2	91 $\pm$ 2	95 $\pm$ 2	90 $\pm$ 2

Full text

1 Introduction

A major application of PA imaging is the monitoring of hemodynamics and changes in sO2 [1], which are indicators of brain activity [2, 3] or injury [4]. In clinical practice, techniques for the monitoring of brain injury can vary widely with the specific application. While there are various techniques that can image hemodynamics, PA imaging can potentially provide better functional information and higher resolution [5]. sO2 is usually calculated via the estimation of abundances of oxygenated and deoxygenated hemoglobin chromophores [6]. In multispectral PA imaging, this concentration estimation is generally done by linear spectral unmixing (SU) [7, 8, 9], which involves solving a set of linear equations for the desired abundances of hemoglobin [10]. While PA imaging is widely used in small animal models with lissencephalic brains, larger and more complex brains remain challenging [5]. The purpose of this work was therefore to investigate the usefulness of multispectral PA imaging in assessing sO2 in a gyrencephalic brain.

2 Material and Methods

To investigate the performance of sO2 estimation by SU in vivo in a neurosurgical setting, we performed PA measurements during an open intervention on a porcine brain, which allowed us to image without the acoustic attenuation of the skull. In this setting, we were also able to take corresponding arterial blood gas (ABG) [11] measurements and thus to compare the quantitative sO2 estimation performance of numerical algorithms for SU [12], against a physiological SaO2 reference value.

Experimental setup

The PA imaging modality used in this study was a custom hybrid PA and US system with a fast-tunable optical parametric oscillator (OPO) laser system (Phocus Mobile, Opotek, Carlsbad, USA) and a 7.5 MHz linear US transducer with 128 elements (L7-Xtech, Vermon, Tours, France), on a DiPhAs ultrasonic research platform (Fraunhofer IBMT, St. Ingbert, Germany) [13]. The custom probe holder covered the transducer with a gold leaf to reduce transducer absorption artifacts. For optimal contrast to noise [14], the PA images were recorded at 760 nm and 858 nm, adding 798 nm as an isosbestic reference.

A porcine brain was continuously imaged for 45 min, as part of an open neurosurgical intervention with our hybrid PA and US probe. Our experiment was carried out following a craniotomy on a three month old female domestic pig. As illustrated in Figure 1a, the probe was fixed over the left hemisphere of the brain to record a sagittal slice, using a gel pad as acoustic coupling. During imaging, the ventilation of the animal was varied and SaO2 and reference measurements with an ABG analyzer were taken. The ventilation changes and reference measurements are detailed in Figure 2 and Table 1. In addition, SaO2 was monitored non-invasively with a pulse oximeter [15] placed on the left earlobe.

Image processing

Using the Medical Interaction Toolkit (MITK) [16] PA image processing plugin [17], the raw PA data was beamformed with delay and sum [18, 19] and von Hann apodization. The resulting images were motion corrected with the corresponding US B-mode images. All input images were averaged over ten recordings per wavelength before linear SU, which was performed on the PA images with five commonly used linear algorithms [20]. To cover a wide range of algorithms, we selected a QR decomposition with Householder transformation [21], a LU (with full pivoting) [22] and a singular value decomposition [23] all from the C++ Eigen [24] library, as well as a weighted [25] (based on QR decomposition) and a non-negative (using least angle regression [26]) least square algorithm both from the C++ Vigra [27] library.

For quantitative validation of the SU algorithms, we selected two regions of interest (ROIs), for which we determined sO2 values. We selected a surface ROI and a deep one to investigate the influence of fluence effects [28] on SU sO2 estimation. We assumed that both ROIs contain arteries, as they had generally high PA signal and distinct characteristic pulsing in the US and PA image streams. The resulting sO2 values for one ROI are the median of all pixels within that ROI that have a higher than noise equivalent total hemoglobin level.

3 Results and Discussion

According to our results in Table 1, the different SU algorithms were similar in estimation performance, with the exception of the non-negative least square boundary effect. While other algorithms can yield physiologically impossible sO2 values (even above 100 %, and theoretically also below 0 %), the non-negativity constraint artificially prevents this. This was especially relevant for the evaluation of changes in ROI 2. All other differences between the algorithms are within their respective standard deviations. The Householder QR algorithm performed the fastest. In the following, we therefore only present the SU results of the QR algorithm. Example slices are shown in Figure 1b&c with the marked ROIs and their corresponding median sO2 value.

Comparing the changes in ventilation with the time course characteristics of the unmixing results of ROI 1 in Figure 2, one can see that linear SU can be used for change detection of sO2 in the cerebral cortex. However, the unmixing results do not closely follow the quantitative values of the ABG reference (Table 1). This illustrates the limits in quantification of sO2 with PA imaging and is even more obvious in deep tissue, considering the small changes in sO2 estimation for ROI 2.

In conclusion, our study suggests that PA imaging can be used to monitor sO2 changes in the cerebral cortex during neurosurgical interventions. However, care must be taken when interpreting sO2 estimation results due to the limits in quantitative accuracy when using linear SU algorithms. This is especially relevant in deep tissue due to fluence dependent spectral coloring. While there are promising approaches to address these fluence effects in general [29, 30, 31] and spectral coloring specifically [28], the translation of quantitative PA imaging deep in tissue remains a major challenge.

ABG arterial blood gas OM Oxygenation measurements under hypoxic conditions ROI region of interest sO2

blood oxygenation SaO2

arterial blood oxygenation rO2

respiratory oxygen SU spectral unmixing PA photoacoustic MITK Medical Interaction Toolkit US ultrasonic PuOx pulse oximetry

Acknowledgements

The authors would like to acknowledge support from the European Union through the ERC starting grant COMBIOSCOPY under the New Horizon Framework Programme grant agreement ERC-2015-StG-37960. The animal experiment was approved by the institutional animal care and use committee in Karlsruhe, Baden-Württemberg, Germany; under Protocol No. 35-9185.81/G-174/16.