A Functional Study of Human Inflammatory Arthritis Using Photoacoustic Imaging

By using our dual-modality system enabling simultaneous real-time ultrasound (US) and photoacoustic (PA) imaging of human peripheral joints, we explored the potential contribution of PA imaging modality to rheumatology clinic. By performing PA imaging at a single laser wavelength, the spatially distributed hemoglobin content reflecting the hyperemia in synovial tissue in metacarpophalangeal (MCP) joints of 16 patients were imaged, and compared to the results from 16 healthy controls. In addition, by performing PA imaging at two laser wavelengths, the spatially distributed hemoglobin oxygenation reflecting the hypoxia in inflammatory joints of 10 patients were imaged, and compared to the results from 10 healthy controls. The statistical analyses of the PA imaging results demonstrated significant differences (p < 0.001) in quantified hemoglobin content and oxygenation between the unequivocally arthritic joints and the normal joints. Increased hyperemia and increased hypoxia, two important physiological biomarkers of synovitis reflecting the increased metabolic demand and the relatively inadequate oxygen delivery in affected synovium, can both be objectively and non-invasively evaluated by PA imaging. The proposed dual-modality system has the potential of providing additional diagnostic information over the traditional US imaging approaches and introducing novel imaging biomarkers for diagnosis and treatment evaluation of inflammatory arthritis.


S1. Selection of wavelengths for photoacoustic imaging (PAI) of arthritis
Before starting this study on human subjects, the optical wavelength for the specific scenario of PAI of human finger joint was optimized based on theoretical analysis and the experiment on a cadaver hand.
The purpose of the experiment is to find the wavelengths that can enable sufficient penetration in the finger join while providing good photoacoustic (PA) imaging contrast between the blood in joint space and the bone within the finger. The optical absorption of both oxygenated and deoxygenated hemoglobin have a predominantly decreasing trend in the visible to near-infrared (NIR) spectral range 1 . The optical illumination at shorter wavelength ensures higher signal-to-noise ratio (SNR) in detecting blood whereas the optical illumination at longer wavelength facilitates deeper penetration in optically scattering soft tissues.  A cadaver hand was used to further identify the optimal wavelength for PAI of arthritis in human peripheral joints. During the experiment, 0.1 ml fresh blood was injected into the metacarpophalangeal (MCP) joints in the cadaver hand under the guidance of ultrasound (US) imaging. All the 5 MCP joints were used. The PA images of each joint were acquired at 570 nm, 580 nm, and 590 nm. Fig. S2 shows the example imaging result from one of the MCP joints. The bone areas were picked up manually based on the parallel US image of the joint, as shown in Fig. S2(a). The needle track during blood injection was observed under the US. The bright pixels around the needle tracks in PA images representing the blood injected were selected, as marked by the red dashed circles in the images. To quantify the enhancement in PA intensity at each wavelength brought by the injected blood, the ratio between the average pixel intensity within the blood injection areas (i.e., the areas marked by the red dashed circles) and the average pixel intensity in a randomly selected area out of the blood injection area, i.e., the area marked by the blue dashed square in Fig. S2(b)-(d), was calculated. The quantified ratios at the three wavelengths (570 nm, 580 nm, and 590 nm) are 2.5, 3.9, and 1.9, respectively. The image depths at all the three wavelengths are sufficient for visualizing the injected blood in the MCP joints. To quantify the contrast of the injected blood over the bone in each PA image, the ratio between the average pixel intensity within the blood injection areas (i.e., the areas marked by the red dashed circles) and the average pixel intensity within the bone areas (i.e., the areas marked by the yellow dashed circle) was calculated. As each wavelength, the ratios from the 5 MCP joints were averaged and the standard deviation was computed. The results for the three wavelengths were compared in Fig. S3. The image contrast between the injected blood over the bone reached a peak at 580-nm wavelength which was then chosen for the study on human subjects involving single-wavelength PAI. (b)-(d) PA images of the cadaver finger joint acquired at 570 nm, 580 nm, and 590 nm, respectively. Guided by US imaging, 0.1 ml fresh blood was injected in the joint region just before the imaging. The areas for blood injection are marked by the red dashed circles. The blue dashed square in (b)-(d) indicates an area out of the blood injection areas to be considered as a background. Figure S3. Statistical analysis in search for the optimal illumination wavelength for PAI of arthritis in human finger joint. The ratios between the average pixel intensity within the blood injection areas and the average pixel intensity within the bone areas were quantified for PA images acquired from the five MCP joints of a cadaver hand. The average and the standard deviation at each of the three wavelengths (i.e., 570 nm, 580 nm, and 590 nm) were calculated for comparison.
For double-wavelength PAI aiming at evaluating hypoxia as the second physiological biomarker of arthritis, the wavelengths of 584 nm and 576 nm, both close to the wavelength of 580 nm determined in the experiment on the cadaver hand, were selected. At 584 nm, the optical absorption coefficients of the oxygenated and deoxygenated hemoglobin are equal (i.e., an isosbestic point). At 576 nm, the contrast between oxygenated hemoglobin over bone has a local peak, as shown in Fig. S1b. PAI of a target joint at these two wavelengths can facilitate the separation of the two forms of hemoglobin as well as the quantification of hemoglobin oxygenation in the joint. We intentionally selected the two optical wavelengths that are close, because largely separated two wavelengths can lead to significantly different optical attenuation in tissue and, therefore, difference in optical penetration. This difference, if not compensated, may affect the accuracy in quantitative evaluation of hemoglobin sO2 by using multiwavelength PAI. In the future, to achieve more accurate sO2 imaging in the joint, the wavelengthdependent light attenuation can be compensated via the established methods 4-6 .

S2. Selection of cutoff threshold to isolate the vasculature in the joint
As mentioned in the manuscript, before a single-wavelength PA image is co-registered with an US image acquired from the same joint, a cutoff at 0.4 is performed to isolate the vasculature and remove the background and the noise in the normalized PA image. Such filtering technique is similar to the wall filter that is used in Doppler US for removing the artifacts due to the motions in background tissue as well as system noise 7 . Similarly, the cutoff threshold adopted in PAI removes the pixels representing background tissue and system noise, retaining only those correlated with the increased hemoglobin content, i.e., hyperemia.
After filtering by performing the cutoff, the number of remaining color pixels in the PA-US combined image, as an example shown in Fig. 2c, quantifies the total area of vascularity within the imaged plane. The optimal cutoff threshold is thereby the value at which the PA images from patients and normal controls have the maximum difference in terms of the vascularity in joint space as reflected by the PA images. Using the data acquired from 16 arthritic joints and 16 normal joints, the cutoff value was searched between 0.2 and 0.7, with a constant step size of 0.1. At each step, the probability (p-value) that the two conditions (arthritis vs. normal) cannot be differentiated based on the density of the pseudo-color pixels in the PA-US combined images was calculated, as shown in Fig. S4. The cutoff value at 0.4 produced the least p-value, i.e., the best differentiation of arthritis joints and normal joints.

S3. Procedure for calculating hemoglobin oxygenation (sO2) in synovium
Before calculating the hemoglobin sO2 image, the synovium segmentation of each joint was conducted based on B-scan US images acquired at the same time of dual-wavelength PAI. The gray scale US image was first denoised using Gaussian filter, as shown in Fig. S5. The denoised image would help in recognizing the boundaries of bones and tendon in the joint. Then the area of synovium, i.e., the region of interest (ROI) to be studied, with its boundary delineated by the metacarpal head, the phalanx and the tendon was segmented. The segmentation results were confirmed by a board certificated radiologist.  Most hyperemia exists in the synovium in the form of a sheet of blood or a blob of blood. Fig. S8 shows the PA-US combined images of 16 arthritic joints. All PA images were acquired at a single-wavelength of the same post-processing procedure as that used in Fig. 2. images were obtained with single -wavelength PAI. And, the 16 healthy joints' images were also collected. In comparison with the sketches in Fig. S7, the form of hyperemia was described in Fig. S8 caption. Figure S7. Sketch of an inflammed human finger joint with hyperemia in synovium.