2.1.

Portable Real-Time Optical Detection Identification and Guidance for Intervention Device

The prototype PRODIGI fluorescence imaging device allows for handheld operation and combines multiband fluorescence emission filters and violet blue light-emitting diodes (LEDs) to enable both WL and AF imaging of tissues. It was validated for the detection and treatment guidance of chronic wound infection, in preclinical and clinical studies.^9,15 Briefly, the device consists of a low-cost consumer-grade sensor-based camera (Model DSC-T900, Sony) housed in a 3-D printed plastic body [Figs. 1(a) and 1(b)]. PRODIGI captures high-resolution 12.1 megapixel color images or 720p videos, which are displayed in real-time to the user in RGB format on a 3.5-in. touch-sensitive LCD screen. PRODIGI’s illumination system consists of two violet-blue LEDs (405 nm, LedEngin LZ4-40UA10, emission 100 mW each for a total irradiance of $0.0044\mathrm{W}/{\mathrm{cm}}^2$ at a distance of 10 cm) and two excitation band pass filters (Thorlabs FB 400–40 nm). During AF imaging, a single dual-band fluorescence filter (Chroma 59022 m, ${\lambda }_{\text{emission}}=500$ to 545 and 600 to 660 nm) housed in a slider is manually moved in front of the camera lens to block excitation light reflected from the tissue and to simultaneously allow the detection of AF from collagen and bacteria. AF images thus consist of green (collagen) and red (bacteria) signals only. PRODIGI is approved for use in clinical trials by Health Canada (ITA application #194991) and by University Health Network (UHN) institutional departments (Reusable Medical Device Committee, Medical Engineering Department).

Fig. 1

Photographs of different views of the handheld PRODIGI prototype device and its sterilization method. (a) The front view shows the main optical components, i.e., the four IR reflective spheres (1), the 405 nm fluorescence light-emitting diodes (LEDs) with excitation filters (2), the emission filter (3), and the filter slide with magnets (4) used for draping. The prototype is powered by a medical grade power supply (5). (b) The back view shows the mechanical pin to slide the emission filter in place during autofluorescence (AF) imaging (6), the heat sink and fans to cool down the LEDs (7) and the camera (8), and a mechanical pin (9) to keep it in place. The device is turned on/off by a rocker switch (10). (c) The device is fully covered by a sterile drape in order to perform intraoperative imaging (11). Four autoclaved magnets (12) keep the drape flush to prevent image quality degradation. (d) IR reflective spheres are used for tracking the device while in use, and steri-strips are tied around each sphere to ensure adequate tracking (13).

2.2.

Optical Tracking System

For this study, PRODIGI was combined with a commercial optical tracking system (OTS) (Polaris, NDI Medical, Waterloo, Ontario, Canada) to track the movement of the device in space relative to the patient over time. The OTS maps the patient in CT space relative to the RT dosimetric plan. This OTS addition consists of an infrared-light camera which tracks four IR reflective spheres (NDI Medical, Waterloo, Ontario, Canada) that are fixed to the external housing of the PRODIGI device [Fig. 1(a)].

The use of the OTS at University Health Network and Mount Sinai Hospital has been previously described.^16–20 Examples of CT-based navigation include guided pelvic osteotomy in sawbones and cadaver models,²⁰ localization of pulmonary nodules under bronchoscopy and fluorescence thoracoscopy in porcine models,¹⁹ critical structure proximity alerts in skull base surgery and endoscopy,¹⁸ and optical-CT fusion to improve superficial disease delineation.¹⁶ CT-based navigation in soft tissue sarcoma is more challenging than other application sites due to soft tissue deformation from patient position differences in RT and surgery, changes in treated tissue throughout RT, and navigation off of historic planning CT due to lack of intraoperative CT in the standard of care. For this study, an infrared-light camera was installed in the treatment room, which enables six degrees of freedom ($x$, $y$, $z$, pitch, yaw, roll) measurements of the camera by localizing the attached IR reflective spheres. Software developed in-house registered and visualized the tracked camera pose relative to previously acquired radiological volumetric image data. In addition to registration, navigation, and covisualization (e.g., visual overlay) of multimodalities, the in-house software platform GTxEyes also performs camera calibration to remove camera distortion and calculates the sensor offset (translation and rotation) to camera center.^16,17 Radiation dose planning information can also be spatially coregistered and displayed on the real and virtual camera views as either isodose lines or colorwash using methods described previously by our group.²¹ Viewing options using the GTxEyes software include orthogonal views through the CT image, corrected PRODIGI image, and a virtual 3-D CT rendered skin surface model viewed from the perspective of a virtual camera at the PRODIGI coordinates.

2.3.

Patient Population

We enrolled 20 adult patients from the Princess Margaret Cancer Centre, University Health Network (Ontario, Canada) sarcoma clinic for treatment with preoperative external beam RT followed by surgical resection of lower limb soft-tissue sarcoma. This feasibility study was approved by UHN’s Research Ethics Board (REB) and Mount Sinai Hospital’s REB (11-1038, 12-0153-E, respectively). Informed consent was obtained according to institutional REBs requirements and good clinical practice (Ref. 22). Patients with pre-existing skin issues who received prior RT or required chemotherapy were ineligible for the study.

2.4.

Imaging Procedures

At the RT planning stage, the planned surgical incision was drawn by the surgeon on the patient’s skin and outlined with a flexible radiopaque wire at the time of the CT simulation scan performed for RT treatment planning purposes [Fig. 2(a)]. This enabled localization of the entire future surgical scar on the 3-D CT rendered skin surface model. CT fiducial markers were also placed on the treatment setup points (small permanent tattoos). These tattoos and CT fiducial markers served as reference coordinates for registration of the PRODIGI and OTS systems in CT coordinates. A commercial extremity immobilization system was used for RT patient positioning.²³ Following CT simulation scan, an appropriate RT plan was designed as per institutional clinical standard guidelines.

Fig. 2

Implementation workflow using the hybrid optically tracked PRODIGI imaging platform. (a) At the planning stage, a computed tomography (CT) scan of the sarcoma site was acquired and fiduciary markers were placed on the six treatment setup points (circled) and planned surgical scar. (b) The optical tracking system (i) and pointer tool (ii) were used to register the permanent tattoos in the CT coordinates and track the PRODIGI camera in space during white light (WL) and AF imaging. Imaging of the sarcoma site was performed during (iii) radiotherapy at fractions 0, 12, and 25 and (iv) during surgery. (c) The GtxEyes software creates a surface model from the CT scans and each WL and AF images are placed at matched camera pose. Each surface model point has the following scalar fields: average WL intensity, maximum AF intensity, and PRODIGI pose.

Imaging was performed throughout sarcoma management [during RT, in the operating room [Figs. 2(b) and 2(c)] and at follow-up appointments]. Imaging was performed at three time points during RT: fractions 0, 12, and 25, i.e., at the beginning, middle, and end of the treatment. Four of the treatment setup points marked with radio-opaque fiduciary markers on the CT simulation scan were used to perform the optical and CT coregistration. An optically-tracked pointer tool using four IR reflective spheres (identical to those fixed to the PRODIGI device) was used to register the patient in space with respect to the IR camera. For this, the pointer was placed sequentially on each tattoo, identified, and spatially registered by the tracking IR camera and visualized in real-time on the CT scans using the software GTxEyes. The locations of the tattoos in the optical tracking coordinate system were then registered to the corresponding points in the CT image using the fiducial markers. Once registration was complete, the planned surgical scar was drawn on the patient’s skin with a marker by superimposing the optically-tracked pointer on the scar visible on the CT scans. An imaging session consisted of both WL and corresponding AF imaging of the planned skin surgical incision and surrounding tissue. Room lights were turned off during AF imaging to avoid background signal and artifacts. The four reflective spheres on the PRODIGI device were pointed toward the IR camera to ensure proper 3-D tracking during the entire session.

The combined WL and AF images were part of a superset of data recorded using the tracking system. The superset also included preoperative patient CT and RT dose volumes. A skin surface model was generated from the patient’s CT, where each surface point held a quantitative tuple that contained RT dose, WL scalar, maximum intensity AF scalar, and camera pose corresponding to the AF scalar [Fig. 2(c)]. Overlay of WL and AF images on the 3-D CT rendered skin surface model was visualized and displayed using the open-source software ParaView (Fig. 3).

To perform imaging with the PRODIGI device in the operating room, a sterilization method was developed (and approved by the institutional Reusable Medical Device Committee) using a commercial grade sterile surgical drape (Cardinal Health Canada, Cat. #29-59029). The elongated drape was used to cover the entire PRODIGI device, including the camera unit and electrical power cord [Figs. 1(c) and 1(d)]. Six small circular neodymium magnets (Super Magnets, 8-mm diameter) were embedded into the slider holding the fluorescence emission filter and another six “complementary” magnets (autoclaved between uses in the operating room) were placed on the outside of the surgical drape to secure it and keep it flat relative to the camera sensor aperture, in order to minimize degradation of image quality. Steri-strips (3M Canada, Cat. #R1547) were applied on the outside of the draped device to secure the drape tightly around the IR reflective spheres in order to insure accurate tracking during imaging. To ensure that this method did not significantly affect the performance of the device, optical resolution was measured using a United States Air Force (USAF) 1951 resolution target consisting of horizontal and vertical lines that vary in size, thickness, and proximity. Clear antiseptic sponges (SoluPrep 3M Canada) were used to sterilize the surgical field in order to avoid the inherent AF signal from the red tint antiseptic normally used for sterilization. Imaging was performed at five time points: before and after sterilization of the surgical site (referred to as OR1 and OR2, respectively), once the surgical flap was raised (OR3), after tumor excision (OR4), and after surgical closure (OR5).

WL and AF imaging were performed on patients at multiple follow-up appointments without the OTS, since CT scans were no longer anatomically representative of the postoperative surgical site anatomy due to inherent deformation of the soft tissue shape and topography following surgery. Patients were tracked for a minimum of three months after surgery, or until closure of the surgical incision, to assess for WCs. A WC was defined as an unhealed surgical incision three months after surgery or as an incision requiring particular wound management, such as surgical debridement, antibiotics and/or vacuum therapy.

2.5.

Microbiology

At each imaging session, swabbing of bacterial red AF positive and negative regions was performed to correlate with standard microbiological laboratory culture testing. Swab cultures were performed at Mount Sinai Hospital’s microbiology department. Light growth of commensal flora was considered negative for the presence of pathogenic bacteria, whereas moderate to heavy growth of commensal flora and of other bacteria species were considered positive for the presence of pathogenic bacteria, and were deemed clinically relevant.¹¹

2.6.

Statistical Analysis

Diagnostic accuracy measures were calculated to determine the ability of the PRODIGI device to correctly identify the presence of pathogenic bacteria (as defined in Sec. 2.4). These measures included accuracy, positive predictive value (PPV), negative predictive value (NPV), and diagnostic odds ratio (DOR), a single indicator of test performance. Accuracy is defined as the number of true positive swabs (TP) plus the number of true negative swabs (TN) divided by the total number of swabs taken. The PPV (the probability that bacteria are present given positive AF) is defined by the number of TP swabs divided by the total number of TP and false positive (FP) swabs. The NPV (the probability that bacteria were not present given negative AF) is defined by the number of TN swabs divided by the total number of TN and false negative (FN) swabs. The DOR here is defined as the ratio of the odds of red AF when pathogenic bacteria is present relative to the odds of red AF when pathogenic bacteria is not present and is calculated by the following formula: $(\mathrm{TP}/\mathrm{FN})÷(\mathrm{FP}/\mathrm{TN})$. The value of the DOR ranges from 0 to infinity with higher values indicating better discriminatory test performance.²⁴

We also examined the clinical utility of PRODIGI to identify patients at risk for developing a WC post radiation treatment using logistic regression analysis to calculate the odds ratio (SAS version 9.4). Here we considered WC as the dependent variable, and AF measured using PRODIGI during radiation treatment (all three fractions combined) as the independent variable. Because some patients had multiple swabs taken at each fraction, we also adjusted for repeat swabs within a patient. Other variables considered as potential confounders for WC included: patient age, gender, ethnicity, tumor size, tumor location, and surgical flap thickness.

3.1.

Longitudinal Imaging During Radiotherapy Treatments

Clinically prescribed RT doses and dose distribution plans could be coregistered and overlaid for visualization of WL, AF, and CT images [Fig. 3(b)]. The clinically approved radiation dose distribution plan displayed on the skin surface images was the dose calculated at 1 mm for the evaluation of direct skin dose or at 3-mm below the skin surface to account for standard radiation dose build-up.²⁵ The accuracy of the registration and dose display was $4\pm 1.6\mathrm{mm}$ on average axially and laterally based on comparison of the fiducial markers and their corresponding skin tattoos on the patient. This error is due to a combination of soft tissue deformation, tumor changes following response to RT, and human variance when positioning the pointer on the skin tattoos. The imaging session added 10 min to the standard RT booking on fractions 0, 12, and 25.

3.2.

Intraoperative Imaging

The optically-tracked PRODIGI platform was used to image the sarcoma site in the operating room to assess for the presence of bacteria at the surgical site. Imaging was performed quickly (10 min) without interference to the regular surgical workflow. Intraoperative WL and AF images taken before and after sterilization of the surgical site (OR1, OR2) were overlaid on the 3-D radiological CT scans to allow comparison to previous preoperative images of the same patient. However, overlay on the CT scans of WL and AF images taken after the surgical incision (OR3, OR4, OR5) were no longer spatially accurate due to tissue deformation at the surgical site.

Sterilization of the device using a commercial surgical drape did not interfere with the operation of the device (including optical tracking) nor did it significantly degrade WL and AF image quality. Indeed, the OTS’s IR camera was able to effectively track the reflective spheres when the drape was fixed tightly around them. WL and AF imaging of a standard 1951 USAF resolution test target using PRODIGI with and without the surgical drape in place demonstrated image resolutions of 111.36 and $31.25\mu \mathrm{m}$. Thus, the use of a surgical drape to ensure sterility of the device during OR use resulted in decreased spatial resolution of WL and AF images. However, this minimal loss of resolution did not adversely affect image quality or interpretation by the clinical user. The use of clear antiseptic sponges did not interfere with the detection of bacterial AF.

3.3.

Postoperative Imaging of Wound Complications

Postoperative WL and AF imaging of the surgical site was performed at follow-up appointments using the PRODIGI device only (no OTS) to assess changes in bacterial burden and monitor healing. Postoperative imaging sessions ranged from 2 days to 10 months after surgery, with an average of four sessions per patient (range: 2 to 9). In total, nine patients (45%) developed WC following surgery. Logistic regression showed that patients who developed WC were almost three times more likely to have red AF regions during radiation treatment (all three fractions combined) compared to patients who did not develop WC ($\text{odds ratio}=2.76$, 95% CI: 0.73, 10.48, $p=0.14$), independent of other clinical risk factors.

3.4.

Bacterial Autofluorescence

Bacterial red AF signal was detected by the device in some patients during RT, in the operating room, and at follow-up appointments. At the RT stage, most patients experienced minor side effects from the radiation, such as mild erythema and dryness of the skin. Some patients ($n=9$) had a pre-existing skin wound during the RT stage caused by a recent diagnostic biopsy that was not completely healed. Positive red fluorescence regions were discrete and found mostly around the wounds and biopsy scars or areas of severe skin erythema (Fig. 4). Imaging before sterilization of the surgical site showed the presence of bacterial red AF signal in three patients. After sterilization of the site, bacterial red AF was still detected in two of these patients, suggesting subsurface localization of bacteria. At follow-up appointments, discrete regions of laboratory culture-confirmed bacterial red AF signal were often found along the incision and surgical staples, while more extensive areas of bacterial red AF were found in patients with WCs.

Microbiological results from the swabbing of red AF positive and negative regions throughout the sarcoma treatment process confirmed the presence of pathogenic bacteria in red AF regions (Fig. 5). In total, 324 swabs were taken during the RT stage, in the operating room, and at follow-up appointments across all study participants. Bacterial species cultured from these swabs included Staphylococcus aureus, Escherichia coli, Streptococcus, Coliform bacilli, and commensal flora. The accuracy of the optically-tracked PRODIGI imaging platform to detect pathogenic bacteria was 85% ($\mathrm{TP}=38$, $\mathrm{TN}=236$). The PPV and NPV were 61% ($\mathrm{TP}=38$, $\mathrm{FP}=24$) and 90% ($\mathrm{TN}=236$, $\mathrm{FN}=26$), respectively. The DOR was 14.4 and was statistically significant (95% CI: 5.8 to 36.0, $p<0.0001$).

Fig. 5

WL and AF imaging throughout sarcoma management procedures and correlative swabbing for a patient with wound complication. (a) WL and AF images at radiotherapy fraction 25 show sarcoma site negative for bacterial red AF signal. Swab #1 confirmed presence of normal commensal flora only. (b) WL and AF images in the operating room before sterilization of the sarcoma site, negative for bacterial red AF signal. AF images are darker due to positioning of the patient. Swab #2 confirmed presence of normal commensal flora only. (c) WL and AF images at follow-up appointment show wound complication six weeks postoperation. Swab #3 taken in a positive red AF area confirmed heavy growth of G Streptococcus. Scale bars: 5 cm.