2.1.

PRODIGI Prototype Imaging Device

PRODIGI consists of a low-cost, consumer-grade, Super HAD™ charge-coupled device (CCD) sensor-based camera (Model DSC-T900, Sony Corp., Tokyo, Japan), with a 35 to 140 mm equivalent $4\times$ zoom lens housed in a plastic body and powered by rechargeable batteries (Fig. 1). It collects high-resolution 12.1 Mpixels color WL and AF images (or videos) in real time ($<1\mathrm{s}$), which are displayed in red-green-blue (RGB) format on a 3.5-in. touch-sensitive color liquid-crystal display (LCD) screen (Fig. 1). Broadband white light-emitting diodes (LEDs), electrically powered by a standard AC125V source, provide illumination during WL imaging, while two monochromatic violet-blue (${\lambda }_{\mathrm{exc}}=405\pm 20\mathrm{nm}$) LED arrays (Model LZ4, LedEngin, San Jose, California) provide 4-W excitation light power during FL imaging (bright, uniform illumination area ${\sim 700\mathrm{cm}}^2$ at 10 cm distance from skin surface). WL and FL images are detected by a high-sensitivity CCD sensor mounted with a dual band FL filter (${\lambda }_{\mathrm{emiss}}=500$ to 550 and 590 to 690 nm) (Chroma Technologies Corp., Bellows Falls, Vermont) in front of the camera lens to block excitation light reflected from the skin surface. Tissue and bacteria AF are spectrally separated by the emission filter and displayed as a composite RGB image without image processing or color-correction, allowing the user to see the bacteria distribution within the anatomical context of the wound and body site. The CCD image sensor is sensitive across ultraviolet ($<400\mathrm{nm}$), visible (400 to 700 nm), and near-infrared (700 to 900 nm) wavelengths to AF of tissues and bacteria, in the absence of exogenous contrast agents. A list of system components for the PRODIGI prototype device is provided in Table 1.

Fig. 1

The handheld prototype PRODIGI imaging device. (a) Front view of PRODIGI showing wound fluorescence (FL) image displayed in real time on the liquid-crystal display screen in high definition. (b) Back view of PRODIGI showing white light (WL) and 405-nm LED arrays providing illumination of the wound, while the FL emission filter is placed in front of the CCD sensor.

Table 1

Main components of the PRODIGI prototype imaging device. The PRODIGI prototype imaging system consists of a Sony DSC-T900 camera, two 405 nm light-emitting diodes, two neutral white LEDs, two switches, four cooling fans, a male connector, a female connector, a DC/DC converter, two heat sinks, a spare Sony lithium-ion battery, an AC/DC power supply, two excitation light source filters, a detection dual bandpass filter, a system shell (built in house), and a carrying case.

Although the current study is a preclinical in vivo validation of PRODIGI in a mouse wound model, PRODIGI is approved by Health Canada as a Class II medical device for clinical investigational testing (Health Canada Investigational Testing Authorization #16114) as well as approvals from our institutional REB (UHN REB #09-0015-A and UHN REB #12-5003) and the Canadian Standards Association. A two-part Phase I, single center, nonrandomized clinical trial of PRODIGI for detecting pathogenic bacteria, guiding wound treatment, and assessing treatment response in chronic wound patients (clinicaltrials.gov identifiers: NCT01378728 and NCT01651845) is ongoing at the University Health Network. A separate paper detailing the results of this clinical trial is forthcoming.

2.2.

PRODIGI Mobile Imaging Device

The PRODIGI mobile device (Fig. 2) is a modified version of the PRODIGI prototype device (Fig. 1), for easier and more convenient use in an animal colony. Although the PRODIGI prototype device is handheld and compact, its uses proved to be cumbersome during FL imaging of the mice within the confined space of the biosafety cabinet in the institution’s animal facility. As such, we developed an iPhone version of PRODIGI, which is a physically different representation of the same technology, operating based on the same optical principles and parameters of FL imaging, and providing the same imaging output as the prototype device described above. A technical comparison of the two devices is provided in Table 2.

Fig. 2

The PRODIGI mobile imaging device. (a) Front view showing the optical components and battery holder of the accessory adapter, which is mounted onto a standard iPhone 4S smart phone. (b) Back view of the device showing the on/off switch and the LCD viewing screen on which the WL and FL images are viewed by the user.

Table 2

Comparison of the PRODIGI prototype versus PRODIGI mobile imaging systems. Both the PRODIGI prototype and PRODIGI mobile systems use the same excitation and emission filters, LED wavelength, and LCD screen. They differ in the image sensor type, physical weight, and dimensions and image spatial resolution. Technically, both devices operate similarly based on autofluorescence imaging achieved using different components.

2.3.

Mouse Skin Wound Model

The procedures described herein have been completed under a specific Animal Use Protocol approved by the University Health Network Animal Care Committee in compliance with all relevant regulatory and institutional agencies, regulations, and guidelines (UHN AUP# 2614.3). Female nude mice aged 8 to 12 weeks (NCRNU-F) were used. Nude mice were chosen for this model because their immune systems are compromised, thereby enabling a robust infection response following inoculation. On the day of surgery (day 0), mice were anesthetized using 2% isofluorane mixed with 100% oxygen. Dorsal skin areas were shaved and cleaned with 70% isopropyl ethanol and 10% povidone-iodine. Two circular full thickness wounds (6-mm diameter) were created on either side of the spine using a biopsy punch with the assistance of scissors and forceps. Mice were subcutaneously inoculated with Gram-positive bioluminescent S. aureus-Xen8.1 from the parental strain S. aureus 8325-4 (Caliper) (${10}^{10}\mathrm{CFU}$) suspended in 30 μL of phosphate-buffered saline (PBS) in one wound and 30 μL PBS in the other wound as a control. Solutions were injected until bubbles formed on the wound surface, and the residual solution was spread on the surface of the infected wound. Gauze and Tegaderm™ were used to cover the wounds after the surgery (Fig. 3). All mice were monitored daily and both WL and FL images were taken of the wounds. Tegaderm™ was changed after each imaging procedure.

Fig. 3

Procedures for creating the mouse skin wound model based on the Guthrie mouse model.⁷ (a) Anesthetized mice receive two equal-sized wounds on both sides of the spine using a biopsy punch. (b) One wound was inoculated with phosphate-buffered saline (PBS) (control; left), the other with Staphylococcus aureus in PBS (right). (c) Tegaderm was applied across the wounds over sterile gauze 15 min after the inoculation. Scale bar: (a–c) 6 mm.

2.4.

White Light and FL Imaging

Regular room lighting was left on during WL imaging of wounds. During FL imaging, the lights were turned off to eliminate artifacts. WL and corresponding FL still images of wounds were collected in a noncontact manner with the device held within 10 to 15 cm from the wound. This distance was kept constant and the image was focused on the center of the wound. The camera’s built-in macromode was used, image acquisition took approximately 1 s, and the images taken were stored for future analysis on the camera’s standard digital memory card. Switching between WL and FL modes was instantaneous using a built in “toggle switch” on the PRODIGI prototype device and by manually turning on and off the FL light source mounted on the top of the iPhone for the PRODIGI mobile device (Fig. 2). Devices were decontaminated between uses with standard 70% ethyl alcohol as an antiseptic wipe.

2.5.

Image Analysis

WL and FL images were transferred to a laptop for image processing. Two programs were used for image processing: MATLAB software (Version 7.9.0, The MathWorks, Natick, Massachusetts) using a custom-written program and ImageJ Software (Version 1.45n). In the MATLAB program, regions of interests (ROIs) were identified from individual $1024\times 1024$ pixel FL images of each wound. RGB images were separated into individual channels. Green (500 to 550 nm emission) and red AF ($>590\mathrm{nm}$) from tissue components and bacteria, respectively, detected by the CCD sensor were naturally aligned spectrally with the red and green filters on the Sony CCD image sensor. Thus, the green and red channels of the RGB image displayed on PRODIGI’s LCD screen were representative of the true tissue and bacterial AF signals detected in vivo. To quantify bacterial levels from individual FL images, the following image processing procedures were used. Briefly, individual green and red image channels from each RGB image were converted to gray scale (the blue channel was not used) and pixels whose gray scale intensity was above a given histogram threshold (selected to reduce the background noise of the raw image) were counted. A red color mask for red FL bacteria was created by finding the local maxima in the color range 100 to 255 gray scale. Then, an inverted green color mask was used to remove the green FL. All pixels with red FL (above the histogram threshold) were binarized and the sum of all “1” pixels was calculated. This was repeated for the green channel of each image. These data gave an estimate of the amount of red bacteria in each image. The number of FL pixels was converted into a more useful pixel area measure (${\mathrm{cm}}^2$) by applying a ruler on the pixel image, thereby providing the total amount of fluorescent bacteria as an area measurement (${\mathrm{cm}}^2$). The sizes of the wounds were traced and measured similarly by converting pixel areas to ${\mathrm{cm}}^2$ of the circled wound area on the WL images. The resolution of the FL images was sufficient to localize bacteria based on regions of FL. ImageJ software was used to separate FL images into red, green, and blue channels using the built-in batch processing function “Split Channels” located within the image menu and color submenu. Each resulting channel was displayed and saved in gray scale. For further analysis, an ROI was identified in each corresponding red, green, and blue channel image. Under the built-in analysis menu, we used the “Set Measurement” function to select and measure the following measurement parameters for each color channel image: pixel area, min. and max. gray scale intensity values, and mean gray intensity values. The average red channel intensity value was determined as (bacterial) FL intensity per square pixel in each red channel image and then used for data analysis and comparison.

2.6.

Tracking Bacterial Load over Time

The mouse skin wound model described above was used to correlate wound status with the progression of bacterial infection ($n=5$; 8 to 12 weeks; NCRNU-F). Daily WL and FL images were taken of the wounds as they became infected over time. Antibacterial treatment (topical Mupirocin three times daily, for a total of 1 day) was applied to the wound site when the red FL intensity peaked. The anti-microbial effect of the treatment was monitored over time using PRODIGI to acquire daily WL and FL images of the wound after treatment. The wounds were monitored for a total of 10 days, after which the mice were sacrificed. Bacterial amounts from FL images and wound size from WL images were quantified using the MATLAB program, and compared over time to determine the wound healing status.

2.7.

Bioluminescence Imaging

BLI was used in the current study to measure the absolute amount of bacteria in vivo, because it is one of the most sensitive and reliable screening tools for determining bacterial load.^12,13 BLI collects the light emitted from the enzymatic reaction of luciferase and luciferin and therefore does not require excitation light.¹⁴ Dinjaski et al. recently showed that near-infrared FL imaging with a dye injection can be an alternative to BLI imaging.¹⁵ In the current study, FL imaging using PRODIGI (without any exogenous FL contrast agent administration) and BLI imaging of inoculated S. aureus bacteria were tracked over time and the FL and BLI intensities were compared ($n=7$). The bacterial BLI signal did not contribute to the FL signal detected by PRODIGI’s consumer grade-CCD camera.

Gram-positive bioluminescent S. aureus-Xen8.1 from the parental strain S. aureus 8325-4 (Caliper) was grown to midexponential phase the day before pathogen inoculation. Bacteria with the BLI cassette produce the luciferase enzyme and its substrate (luciferin), thereby emitting a 440 to 490 nm bioluminescent signal when metabolically active^16,17 (Fig. 4). The bacteria (${10}^{10}\mathrm{CFU}$) were suspended in 0.5 mL of PBS and injected into the wounds of female athymic nude mice ($n=7$; 8 to 12 weeks; NCRNU-F Homozygous). To detect S. aureus bioluminescence, BLI images of the wound were acquired before, immediately after, and 1, 2, 3, 4, 5, 6, and 7 days postinoculation inside the dark chamber of the Xenogen IVIS Spectrum Imaging System 100 (Caliper, Massachusetts), using an exposure time of 10 s. The mice were kept under gas anesthesia consisting of 100% oxygen mixed with 2% isoflurane during imaging. BLI images were captured using Living Image In Vivo Imaging software (Caliper, Massachusetts). ROIs were digitally circumscribed over the wound and the total luminescence intensity counts were measured within the ROIs for each time point imaged. The absolute amount of bacteria measured from the BLI signals was tested for correlation with the corresponding FL signals on the FL images taken over time of the same wound using PRODIGI (as described above).

Fig. 4

Bioluminescence imaging of bacteria in vivo. Preliminary data showing the S. aureus-Xen8.1 is both autofluorescent [red color with PRODIGI (a)] and bioluminescent [with IVIS imaging system (b)] when grown in our mouse skin wound model. Scale bar: (a and b) 1 cm.

2.8.

Statistical Analysis

Linear correlation analysis was performed to determine the level of correlation between the longitudinal FL with BLI imaging intensity measurements using the Pearson’s correlation test. Statistical analysis was carried out using statistical analysis system (SAS) v 9.2.

3.1.

Tracking Bacterial Load over Time

The red FL signal was first detected on day $3\pm 1$ (range 2 to 4) and peaked on day $5\pm 1$ (range 4 to 6). Two mice had serious cross-contamination of the adjacent wounds and were sacrificed. In the remaining mice ($n=3$), red FL was observed in the wounds that were inoculated with S. aureus but not in the adjacent control wounds where only PBS was injected.

Representative WL and FL images for a single mouse tracked over 10 days are shown in Fig. 5. Bacteria (red FL) were not visible on days 1 and 2 postinoculation, but as the bacteria proliferated and colonized the wound, red FL was visualized on day 3 and increased over days 4 to 6, by which time the infection could be visualized under WL. On day 7, the red FL intensity peaked and then began to drop. Mupirocin was applied to the wound on day 7 and the bacterial load, visualized as red FL, decreased significantly, to almost zero intensity on day 8. However, red FL reappeared on days 9 and 10.

Fig. 5

Representative WL and FL images for a single mouse tracked over 10 days. (a) PRODIGI prototype images showing the two equal-sized wounds on both sides of the spine. The right wound was inoculated with S. aureus in PBS and the left wound was inoculated with PBS only (control). The top row shows WL images, the middle row shows FL images, and the bottom row shows MATLAB quantified images, corresponding to bacterial areas and intensities. The FL imaging data demonstrated a significant increase in bacterial FL intensity in the wound inoculated with S. aureus, compared with the control wound, peaking on day 6. Mupirocin (day 7, red arrow) significantly decreased bacterial FL on day 8 to almost zero, indicating treatment effect. Bacteria increased again on days 9 and 10. (b) Graph showing quantitative changes in bacterial load from FL images obtained in panel (a).

3.2.

Correlating FL with BLI

A significant positive correlation was observed between the bacterial FL captured by PRODIGI and the absolute bacterial load measured by the Xenogen imaging system (Pearson correlation coefficient 0.6889, $p$-value 0.04; Fig. 6). Both imaging modalities showed an increase in bacterial load from day 0 to day 4, with a drop in bacterial load after day 4. Overall, the changes in FL intensity of the bacteria correlated with the bacterial BLI intensity (e.g., absolute bacterial load) as the bacterial amount increased and subsequently decreased in the wounds. A representative example of a mouse tracked over time using WL, FL, and BLI imaging is shown in Fig. 6. FL intensity was calculated from the red channel of RGB images using ImageJ software.

Fig. 6

Preclinical data show that pathogenic bacterial autofluorescence (AF) intensity correlates with bacterial load in vivo. (a) Time course PRODIGI mobile images of skin wounds in a mouse prior to and after inoculation with bioluminescent S. aureus-Xen8.1 (${10}^{10}\mathrm{CFU}$ in 30 μL PBS). Representative WL (top row), AF (middle row), and bioluminescence (bottom row) images are shown for each time point to 7 days after inoculation in a wounded mouse. BLI imaging gives absolute bacterial amount in vivo. Red arrows show when the tegaderm bandage was exchanged, causing some bacteria to be removed from the surface. (b) Average red FL from S. aureus-Xen8.1 ($n=7$ nude mice) shown as a function of time demonstrating an increase in daily S. aureus bacterial FL (calculated from red channel of RGB images using ImageJ software). At days 2 and 7, tegaderm bandages were exchanged as per animal protocol. Average bacterial FL peaked at day 4 postinoculation. (c) Corresponding time course bioluminescence data (calculated from ROI) show similar increase and peaking at day 4 in total bacterial load in the wound. Data indicate strong positive correlation (Pearson correlation coefficient $r=0.6889$) between total bacterial AF in a wound and the bacterial load in vivo. Standard errors are shown. Scale bars: (a) WL 1.5 cm and AF, BLI 1 cm.