Fluorescence guidance improves the accuracy of radiological imaging-guided surgical navigation

Abstract

Background:

Imaging-based navigation technologies require static referencing between the target anatomy and the optical sensors. Imaging-based navigation is therefore well suited to operations involving bony anatomy; however, these technologies have not translated to soft-tissue surgery. We sought to determine if fluorescence imaging complement conventional, radiological imaging-based navigation to guide the dissection of soft-tissue phantom tumors.

Methods:

Using a human tissue-simulating model, we created tumor phantoms with physiologically accurate optical density and contrast concentrations. Phantoms were dissected using all possible combinations of computed tomography (CT), magnetic resonance, and fluorescence imaging; controls were included. The data were margin accuracy, margin status, tumor spatial alignment, and dissection duration.

Results:

Margin accuracy was higher for combined navigation modalities compared to individual navigation modalities, and accuracy was highest with combined CT and fluorescence navigation (p = 0.045). Margin status improved with combined CT and fluorescence imaging.

Conclusions:

At present, imaging-based navigation has limited application in guiding soft-tissue tumor operations due to its inability to compensate for positional changes during surgery. This study indicates that fluorescence guidance enhances the accuracy of imaging-based navigation and may be best viewed as a synergistic technology, rather than a competing one.

1 |. INTRODUCTION

Soft-tissue sarcomas are cancers originating from mesoderm-derived connective tissues. They demonstrate generally poor responses to chemotherapy and modest responses to radiation; complete surgical excision of a non-metastatic primary tumor is the only reliable curative treatment.^1–4 Soft-tissue sarcomas are optimally removed with the primary tumor surrounded by a margin of normal, noncancerous tissue—a surgical technique termed wide-local-excision (WLE).^5,6 A successful WLE includes the tumor and the tumor’s reactive zone and demonstrates no evidence of tumors at the specimen’s periphery upon final histological review, a finding termed a “negative” margin. Despite the importance of complete tumor removal, there have been no substantial advancements in surgical technique for soft-tissue sarcomas since the adoption of cross-sectional imaging, and the rates of unsuccessful WLE for soft-tissue sarcoma are 15%–23% in the largest reports.^7,8

Surgical navigation is a broadly defined group of technologies that seek to improve operative safety and efficacy by enhancing surgeons’ ability to visualize—directly or indirectly—important anatomical structures that the surgeon desires to remain uninjured (e.g., motor nerves) or removes entirely (e.g., cancers).⁹ To date, orthopedic surgeons have primarily investigated anatomic imaging-based navigation using computed tomography (CT). CT- and magnetic resonance imaging (MRI)-based navigation has advantages when used for skeletal surgery where the navigational reference is anchored to the surgical target—pelvic bone sarcomas—or the operative field is relatively static—spine surgery^10–12; however, neither technology adapts to positional changes. For this reason, neither CT-nor MRI-based navigation is considered an accurate guide for soft-tissue tumor surgery.¹³

Fluorescence-guided surgery (FGS) is a nascent technology that seeks to improve the safety and effectiveness of surgery through the use of targeted fluorescent reporters—fluorophores—that identify tissues of interest.¹⁴ Because it images—in real time—molecules bound to target tissues, FGS provides dynamic feedback to surgeons that is independent of preoperative imaging or positioning. One major limitation of FGS is that tissue penetration, even for fluorophores with near-infrared emission peaks, is <2 cm. Therefore, to date, FGS has been applied primarily to cancers that are superficial or removed piecemeal.^15–17 In prior work we demonstrated successful WLE of phantom tumors using near-infrared, subsurface FGS alone; however, margin depth varied widely.¹⁸ In this project we sought to evaluate the role of anatomic imaging-based navigation to complement FGS for tumors requiring WLE. We performed, using a validated and highly reproducible human connective tissue-simulating phantom model, iterative dissections of soft tissue masses using combinations of CT-, MRI-, and fluorescence-based navigation.^18,19 We hypothesized that the highest accuracy for mass excision would occur with combined CT, MRI, and FGS technologies.

2 |. MATERIALS AND METHODS

2.1 |. Regulatory approval

Because this study used gelatin phantoms—nonhuman, nonanimal—no regulatory approval was required.

2.2 |. Phantoms

The validated gelatin-based phantom model used in this study was an ideal instrument for studying the effectiveness of surgical imaging technologies, particularly technology indicated for improving the WLE technique.^18,19 Gelatin phantoms were opaque, thereby obscuring the location of each simulated “tumor” under typical white light illumination (i.e., room light). Furthermore, the gelatin phantoms offered no ability to use tactile sensation to delineate hidden structures; therefore, the localization of a structure enclosed in the gelatin depended solely on the imaging technology under evaluation.

The phantom model was a volume of semi-solid material with optical properties identical to muscle or adipose—simulating a tumor’s investing stroma—into which an 8-cm³ inclusion simulating the tumor was poured (Figure 1).¹⁸ The stroma was created first using 10% weight-per-volume (w/v) gelatin (G2500, Sigma-Aldrich) mixed with ~70°C tap water and stirred vigorously while cooling to ~40°C. Once cooled, whole bovine blood (Lampire Biological Laboratories, Inc.) and intralipids (Baxter Healthcare Corporation) were added to the gelatin mixture in different amounts depending on the tissue type being simulated: adipose-simulating stroma was created using 0.5% volume-per-volume (v/v) whole bovine blood and 0.75% v/v intralipid. Muscle-simulating stroma was created using 2% v/v whole bovine blood and 1% v/v intralipid.

FIGURE 1.

Demonstration of gelatin phantom structure with 8-cm³ cubical gelatin “tumor” inclusion hidden within the cylindrical gelatin stroma

Stroma mixtures were poured into 500 ml cylindrical molds (Nalgene) pretreated with aerosolized canola oil (Food Club, Topco Associates LLC). After pouring, an acrylic bar with a cross-sectional area of 4 cm², treated with canola oil, was suspended in the stroma. The acrylic bar created a void into which the tumor inclusion material (8 cm³) was later poured. Ten acrylic bars were available, each capable of producing a void between 2 and 6 cm deep. The acrylic bars were placed at random within the 500 ml molds of stromal material. The 500 ml molds, with acrylic bars, were then moved to a cold storage room (4°C) until the stromal material congealed (~1 h). Once congealed, the acrylic bars were removed, and the molds were ready for injection of inclusion material (Supporting Information: Figure A1).

Inclusions were made identically, regardless of intended imaging. Inclusions were made with the same process as the stroma but with 10% w/v gelatin, 1% v/v whole bovine blood, and 1% v/v intralipid. To facilitate CT- and MRI-based navigation, each inclusion received a volume-equivalent human dose of iohexol (1.25 ml/kg, GE-Healthcare Inc.) and gadoterate (0.2 ml/kg, Guerbet-LLC). To facilitate fluorescence imaging, inclusions received scaled amounts of IRDye 800 CW carboxylate (LI-COR-Biosciences, Inc.). All contrast doses were scaled from human adult blood volumes, assuming high vascular density common to high-grade soft-tissue sarcomas.^20,21 IRDye 800 CW carboxylate was administered to each inclusion to obtain a 4:1 ratio relative to the phantom stroma, which replicates the adipose and muscle-to-tumor ratios we have measured in human xenografts and in vivo human work for ABY-029.²² After the inclusion material congealed (~1 h), stroma-simulating material was poured into the inclusion void until it was filled.

2.3 |. Study design

Ten phantoms, five adipose calibrated and five muscle calibrated, were assigned to each of nine groups (90 phantoms total). Eight groups were assigned one or more navigation modalities; one group had no navigation or imaging whatsoever (Table 1). The sample size was based on the primary objective: to compare the accuracy of subsurface FGS with and without CT and/or MR navigation (CT-N, MR-N) to perform WLE of a phantom tumor with 1-cm margins. Each dissected phantom yielded one data set: surface fluorescence at each margin surface (6 measurements/phantom, 540 measurements total), margin accuracy (deviation from 1- cm goal margin, 540 measurements), WLE success or failure (i.e., positive/negative margins, 90 measurements total), specimen tumor alignment (90 measurements), and dissection duration (90 measurements).

TABLE 1.

Aggregated accuracy-related performance results for all navigation cohorts

						Margin accuracy (deviation from 1 cm goal]					WLE result		Tumor-specimen alignment
Results by groupa FGS	FGS	CT	MRI	CT-N	MR-N	Mean Δ (mm)	SD (mm)	p Value	Max	Min	Positive margins (n)	p Value	(deg.)	Std Dev	p Value	Max	Min
Group 0 (no imaging)	0	0	0	0	0	8.1	6.1	<0.001	27.9	0.0	15		23.2	12.8	<0.001	44	4.0
Group 1 (ref)	0	1	1	0	0	2.2	1.7		5.4	0.1	0		9.4	7.9		20	0.0
Group 2	0	0	0	1	0	3.4	2.6	0.033	9.0	0.0	0		6.4	5.4	0.410	15	1.0
Group 3	0	0	0	0	1	4.4	2.9	<0.001	10.0	0.0	4		8.3	9.0	0.736	29	0.0
Group 4	0	0	0	1	1	3.1	2.8	0.103	13.1	0.1	1		3.7	3.9	0.084	10	0.0
Group 5	1	1	1	0	0	3.0	2.3	0.139	11.3	0.3	0		7.2	8.0	0.501	28	0.0
Group 6	1	0	0	1	0	1.9	1.5	0.523	5.2	0.0	0		3.4	3.5	0.069	9	0.0
Group 7	1	0	0	0	1	2.1	1.9	0.780	10.0	0.1	1		5.4	5.0	0.223	16	0.0
Group 8	1	0	0	1	1	2.0	2.0	0.919	10.0	0.1	1		4.2	3.7	0.114	10	0.0
Fluorescence
FGS - no (ref)	0					3.2	2.6		9.4	0.1	5		7.0	6.6		19	0.0
FGS - yes	1					2.3	2.0	<0.001	10.9	0.1	2	0.45	6.4	6.6	0.005	21	0.0
All imaging-based navigation
Navigation - no (ref)				0	0	2.6	2.0		8.4	0.2	0		8.3	8.0		24	0.0
Navigation - yes				1	1	2.8	2.4	0.004	9.6	0.1	7	0.015	5.2	5.1	0.006	15	0.0
CT imaging-based navigation
CT navigation - no (ref)				0		2.9	2.3		9.2	0.1	5		7.6	7.5		23	0.0
CT navigation - yes				1		2.6	2.3	0.004	9.3	0.1	2	0.45	4.4	4.1	0.001	11	0.0
MR imaging-based navigation
MR navigation - no (ref)					0	2.6	2.1		7.7	0.1	0		6.6	6.2		18	0.0
MR navigation - yes					1	2.9	2.5	<0.001	10.8	0.1	7	0.015	5.4	5.4	0.014	16	0.0
Fluorescence + CT navigation
All others (ref)						2.9	2.4		5.2	0.0	7		3.4	3.5		9	0.0
FGS+ CT navigation	1	0	0	1	0	1.9	1.6	0.045	9.8	0.1	0	0.61	6.4	6.2	0.082	18	0.2

^aCT, computed tomography; CT-N, computed tomography-guided navigation; FGS, fluorescence-guided surgery; MRI, magnetic resonance imaging; MRI-N, magnetic resonance imaging-guided navigation.

Assuming a Type I error rate of 0.05, 60 fluorescence and 60 margin depth measurements per group provided sufficient power to examine the difference in margin between all experimental conditions and the absolute control (1−β_range = 0.99–0.99). Sufficient power was observed to detect the difference between combined individual imaging modalities and combinations of modalities for all cases (1−β_range = 0.80–0.99) except fluorescence/CT + MRI (1−β = 0.07) and fluorescence/fluorescence + MRI + CT (1−β = 0.69). There was insufficient power to detect a difference between conditions with combined imaging modalities (1−β_range = 0.15–0.65).

2.4 |. Imaging-based navigation

In preparation for imaging, custom imaging platforms were constructed from acrylic. Platforms were 25 × 25 cm² and could accommodate four phantoms (Supporting Information: Figure A2). Each platform was equipped with 10 multi-modality fiducial markers (IZI-Medical-Products), 5 of which were out-of-plane, mounted on an 18-cm height acrylic bar in the platform’s center. Following the arrangement of the phantoms, all platforms were imaged via CT (Definition-AS 64, Siemens) and MRI (Skyra-3T, Siemens). For groups 2–4 and 6–8, CT and MR images were loaded into a StealthStation-sX surgical-navigation system (Medtronic). Groups 1 and 5 had standard images (axial, coronal, and sagittal) available at the time of dissection without navigation (navigation controls); group 0 had no imaging (absolute control).

2.5 |. Fluorescence-based navigation

Real-time fluorescence imaging for groups 5–8 was performed with an open-air Solaris imaging system (PerkinElmer), set to its 750 nm channel. During dissection, a square region of interest of 4 cm² was used to measure average fluorescence values in real time.

2.6 |. Phantom dissection

One investigator (ERH), trained in StealthStation and Solaris use, performed all dissections (Figure 2). Phantom dissections occurred group by group. Groups were presented in random order, and the dissector was blinded to both the location and depth of the inclusions. Per StealthStation workflow, the platform fiducials were verified with the system’s optical probe. The dissector was presented with the optical probe, a #10-scalpel, and each phantom dissection was timed. The dissector was instructed to remove the cubical inclusion—8 cm³—with a 1 cm stromal margin, thereby producing a 4 cm/side cube (Figure 3). The dissector was instructed to attempt to have the faces of the cubical inclusion oriented in parallel to the faces of the resected specimen; this metric was recorded as tumor-specimen alignment. The data resulting from the dissection process included the absolute value of margin difference from the 1-cm goal (Δ, mm), margin status (negative/positive), specimen-tumor alignment (°), and dissection time (s) (Supporting Information: Figure A3).

FIGURE 2.

Demonstration of phantom dissection workflow, with undissected phantoms, mounted on imaging platforms, with imaging-based and fluorescence-based navigation technologies available (A). Photograph of StealthStation screen demonstrating CT navigation of gelatin phantom (B).

FIGURE 3.

Tissue phantom before (A) and after dissection (B and C) as viewed with white light (B) and fluorescence (Solaris) (C) guidance; inclusion was dissected successfully with negative margins through indirect fluorescence visualization alone with no visual or tactile guidance. Permission obtained from Sage Publishing (image previously published in Cancer Control, 25(1) 2018).

2.7 |. Statistical analysis

Descriptive statistics across all experimental groups, including mean (M), standard deviation (SD), and counts were reported for margin accuracy, margin status, incision tumor alignment, and dissection duration outcomes.

Univariate and multivariate regression was utilized to explore the relationship between different imaging—modalities and margin accuracy, specimen tumor alignment, and dissection duration outcomes. Moreover, interaction terms were included in the multivariate models to investigate the combined effect of multiple imaging technologies on these outcomes. Further regression analyses comparing the performance of fluorescence imaging for margin accuracy, specimen tumor alignment, and dissection duration outcomes across the optical density of the stroma (adipose vs. muscle) were evaluated through the introduction of an interaction term between the fluorescence imaging condition and stroma.

2.8 |. Funding

This investigation was funded by the National Institutes of Health/National Institute of Biomedical Imaging and Bioengineering (NIH/NIBIB).

3 |. RESULTS

3.1 |. Margin accuracy

Margin accuracy was improved significantly for fluorescence- and CT-based navigation technologies compared to no navigation (Table 1). The most accurate dissections were yielded with combined CT and fluorescence navigation (M Δ = 1.9 mm, SD = 1.6 mm) as compared to any alternative single or combination of navigation technologies (Intercept = 4.9 mm, B = −2.8 mm, p = 0.045; Figure 4); the least accurate dissections were yielded by the absolute control (M Δ = 8.2 mm, SD = 5.9 mm). CT and fluorescence navigation also yielded the highest margin precision (Figure 5). Univariate analysis demonstrated the accuracy of each technology with FGS resulting in the greatest reduction in M Δ (Intercept = 4.2 mm, B = −1.9 mm, p < 0.001), followed by CT (Intercept = 3.9 mm, B = −1.3 mm, p < 0.001), and MRI (Intercept = 3.7 mm, B = −0.85 mm, p = 0.004; Supporting Information: Table A1). Fluorescence imaging demonstrated improved margin accuracy irrespective of its paired imaging modality (Figure 6).

FIGURE 4.

Violin plot demonstrating margin accuracy for all imaging-defined phantom groups, with subdivision according to the use of CT navigation and color-coded to indicate the use of fluorescence imaging (blue)

FIGURE 5.

Box plot demonstrating aggregated margin accuracy with subdivision according to the use of fluorescence and color-coded to indicate the use of CT navigation (blue)

FIGURE 6.

Violin plot demonstrating margin accuracy for all imaging-defined groups, with subdivision according to phantom stroma optical properties (fat or muscle) and color-coded to indicate the use of fluorescence imaging (blue)

3.2 |. Wide local excision success

WLE success, defined as having no inclusion material at the specimen surface, was greatest for the following conditions: standard image alone, CT scan alone, fluorescence guidance alone, and the combination of fluorescence and CT (Table 1). The lowest excision success resulted from absolute control with 15 positive margins. Aside from the absolute control group, all positive margins occurred with MR-N (p = 0.015); no positive margins occurred with CT-N; however, this study was not powered to support this finding.

3.3 |. Specimen-tumor alignment

The absolute control yielded the worst alignment (M = 23, SD = 13). Univariate analysis demonstrated the accuracy of each technology (Table 1) with CT imaging resulting in the greatest alignment (Intercept = 11°, B = −6.4°, p < 0.001) followed by fluorescence imaging (Intercept = 11°, B = −5.4°, p < 0.001). Specimen-tumor alignment was the highest—least rotational deviation—for the combination of CT and fluorescence navigation technologies (M = 3.4°, SD = 3.5°) as compared to any alternative single or combination of navigation technologies (Intercept = 13.6°, B = −9.8°, p < 0.001; Supporting Information: Table A2, Figure 7).

FIGURE 7.

Violin plot demonstrating margin accuracy for all imaging-defined phantom groups, with subdivision according to the use of CT navigation and color-coded to indicate the use of fluorescence imaging (blue)

3.4 |. Dissection duration

Dissection duration was longest for the combined CT and fluorescence navigation (M = 455 s, SD = 54 s); the shortest duration was found for standard imaging (M = 149 s, SD = 49 s, Table 2). Univariate analysis demonstrates that fluorescence imaging resulted in the greatest increase in dissection duration (Intercept = 232 s, B = 180 s, p < 0.001) followed by CT imaging (Intercept = 286 s, B = 70 s, p = 0.005; Supporting Information: Table A3).

TABLE 2.

Modality-specific dissection duration

	Duration
Results by group	Mean (m:s)	Std Dev (s)	p Value
Group 0 (no imaging)	NA	NA
Group 1 (ref)	2:29	49.4
Group 2	4:52	44.7	<0.001
Group 3	3:47	54.8	0.003
Group 4	4:32	41.5	<0.001
Group 5	6:06	76.3	<0.001
Group 6	7:35	53.6	<0.001
Group 7	6:43	58.0	<0.001
Group 8	6:56	61.0	<0.001
Fluorescence
FGS – no (ref)	3:55	71.0
FGS – yes	5:44	68.1	<0.001
All imaging-based navigation
Navigation – no (ref)	4:17	127.9
Navigation – yes	5:44	100.3	0.896
CT imaging-based navigation
CT navigation – no (ref)	4:46	119.6
CT navigation – yes	5:59	95.5	0.005
MR imaging-based navigation
MR navigation – no (ref)	5:16	125.8
MR navigation – yes	5:30	100.1	0.670
Fluorescence + CT navigation
All others (ref)	5:04	44.7
FGS + CT navigation	7:35	119.4	0.145

Abbreviations: CT, computed tomography; FGS, fluorescence-guided surgery; MR, magnetic resonance.

3.5 |. Stroma optical density – Fat/muscle

The performance of fluorescence imaging did not differ significantly for fat and muscle stromal optical densities (Table 3). Specifically, differences in margin accuracy (Intercept = 4.5 mm, B = 0.7 mm, p = 0.227), specimen-tumor alignment (Intercept = 11°, B = −0.4°, p = 0.913), and dissection duration (Intercept = 219 s, B = −9.8 s, p = 0.0754) were not significantly different (Supporting Information: Table A4).

TABLE 3.

Margin, alignment, and timing difference estimates across stroma utilizing multivariate OLS regression

	Fat	Muscle
Margin Δ (mm)	−2.26*	−1.58*
Margin estimate (mm)	2.2	2.38
Specimen-tumor alignment Δ (deg.)	−5.2	−5.62
Specimen-tumor alignment estimate (deg.)	5.8	4.3
Time Δ (s)	182.8*	173.01*
Time estimate (s)	401.45	417.72

Note: Compared to the absence of any imaging technique.

^*p < 0.05.

4 |. DISCUSSION

Anatomic-based navigation for bony tumors has clear benefit, while enhanced localization of soft tissue tumors has proven more challenging. High variability in tumor location, size, and surgical approach requires initial evaluation of imaging modalities using phantom models. To this end, we assessed fluorescence guidance combined with anatomic-based surgical navigation to determine if this combination improves dissection metrics compared to conventional imaging, imaging-based navigation, or fluorescence-based navigation alone using a validated soft-tissue sarcoma phantom model. Our bivariate results indicate that there is positive synergy with the addition of fluorescence guidance to CT or MRI navigation, with significant improvements in margin accuracy over these technologies applied individually; the highest accuracy was achieved with CT-N with fluorescence. When combined with fluorescence, fused CT + MRI imaging showed no advantage over CT or MRI alone; CT + fluorescence performed best of the three combinations. We attribute this finding to the sharper delineation between stroma and inclusion with CT compared to MRI. Also, we found incomplete agreement between CT and MRI navigated images, causing some uncertainty about inclusion location. In light of the higher positive margin rate with MRI compared to CT, we believe CT navigation to be a more reliable indicator of location. LE’s success and tumor-specimen alignment also showed significant improvements with combined fluorescence and imaging navigation. Dissection duration increased with the use of with combined-fluorescence and conventional imaging navigation, the only disadvantageous finding associated with fluorescence.

The ability to derive clinical conclusions from this gelatin phantom model will assuredly draw skepticism and therefore we address it immediately, separate from the standard “limitations” paragraph. We acknowledge fully that from a tactile and physical appearance perspective, the gelatin phantoms do not—and were not intended—to simulate human surgery. What this phantom model facilitates is the complete blinding of the location of a “tumor” without any ability of the “surgeon” to visualize, palpate, or in any way discriminate it from the investing tissue. Therefore, this model provides a pure test of navigational technologies, in comparison to an animal model whereby the tumor may be readily palpable; this is analogous to a pilot flying “on-instruments” compared to “visual-flight-rules.”²³ Second, the gelatin phantom is highly reproducible and cost-effective, affording the investigative team the ability to present to the “surgeon” iterative circumstances that allow discrete methodological or technical alterations and therefore is conducive to controlled, hypothesis-driven research. Third, the gelatin phantom provides a vehicle to explore these technologies with no risk to patients and without unnecessary animal use. A true “on-instruments” clinical test of navigational technologies would be unethical, exposing patients to the morbidity of injury, positive tumor margins, or death. Gelatin phantoms allow testing to be conducted without fear of harm and therefore facilitate an unalloyed test of the technology. However, we acknowledge that the phantom model has limitations. At present, it does not allow for the tumor to be surrounded by multiple tissue types, as soft-tissue sarcomas are commonly invested in adipose, muscle, fascia, and other tissues. The phantoms do not have the character or consistency of the tissues they optically duplicate, and therefore the experience of dissecting them is not obviously transferrable to surgical skill development.

This study has additional limitations that warrant discussion. The doses of iohexol and gadoterate provided in the phantom inclusions were scaled based on an average adult’s blood volume, assuming that the tumor was highly vascular; however, this is not uniformly true. Another potential limitation of the phantom model is that it assumes homogeneous fluorophore uptake within the tumor inclusion. While our work with small human soft-tissue sarcoma xenografts in a murine model shows relatively uniform fluorescence intensity, our first-in-human work with an epidermal growth factor receptor-targeting fluorophore shows strong intensity in areas of viable tumor, but the weaker intensity in necrotic and nonviable areas that occur frequently in larger tumors. Using a dual-agent approach, with targeted and nontargeted fluorophores, we improved whole-tumor fluorescence for soft-tissue sarcomas; thus, we do not believe that the current model is flawed. However, the current model may be altered to incorporate signal voids.²⁴

Imaging-based surgical navigation has existed for >20 years and has been investigated heavily in orthopedic oncology, primarily for WLE of primary bone cancers and particularly about the pelvis.^{10,13,25–44} Intralesional applications of navigation have been reported.^30,45 Advantageous clinical findings associated with imaging-based oncological navigation include more accurate margins, more accurate reconstructions, and lower disease recurrence.^{10,13,27,30–32,34,39,44} Navigated-WLE for soft-tissue sarcomas has been described, however, it has been described almost universally in the setting of tumors with bone involvement, with navigation applied to the bone resection.^32,34,40,44

Navigating soft-tissue masses remains a major limitation of imaging-based navigation. As described by Wong et al., “Following-skin-incision-and-surgical manipulation, soft-tissues will deform, and the real-time anatomic location will change from that on preoperative virtual images. Only the bony anatomy remains the same… soft tissue resection still requires a conventional surgical technique.”¹³ In the only series of human soft-tissue sarcomas to undergo attempted CT-guided navigation, Bosma et al. reported no improvement in WLE success.²⁷ Reijers et al. reported probe-based electromagnetic navigation for soft-tissue sarcomas of the abdomen, however, lessons were limited due to sample size.⁴⁶

A precedent exists for troubleshooting navigation systems using simulated tumor models. Navigated resection of bone tumors, created from polymethylmethacrylate in pigs, was reported by Cho et al.²⁹; Zamora et al. reported similar work in cadavers.⁴⁷ Eccles et al. simulated soft-tissue sarcoma surgery in a cadaver using silicone caulk injected into balloons.⁴⁸ Using MRI-based navigation, they demonstrated successful dissection of 98 simulated tumors with a high agreement in margin depth and accurate tumor removal. While our model does not resemble human anatomy as a cadaver does, it facilitates no visual or tactile hints about the location of the “tumor,” thereby providing a more rigorous evaluation of the navigation technology.

The experiential lessons from this study indicate that achieving an improved outcome when combining navigational tools will require an ordinal surgical workflow. CT- and MRI-based navigation facilitated rapid identification of the hidden masses, and therefore may be best used initially as tools for macroscopic localization. In comparison, the initial tumor location with fluorescence is slow and offers less initial confidence.¹⁸ The benefits of fluorescence became more apparent after the initial identification of the mass and the dissection neared its endpoint. In these final steps of the dissection, fluorescence measurement allowed dissection to proceed with continuous, real-time, and quantitative feedback, particularly once the specimen became mobile and the validity of the imaging-based navigation was uncertain.

Dissection duration was significantly longer when using fluorescence guidance compared to radiological imaging. This was due partially to the Solaris workflow, which requires several seconds to ascertain fluorescence values each time the specimen is moved; however, our experience is that other imagers of its vintage have similar imaging times. Newer imagers, however, appear to offer real-time fluorescence measurement.

5 |. CONCLUSION

Despite positive improvements in bone-tumor resections, in its current form imaging-based navigation has limited application in guiding soft-tissue tumor surgeries due to its inability to compensate for the positional change. This study indicates that fluorescence guidance provides real-time assistance for the localization of hidden anatomical targets, enhances the accuracy of imaging-based navigation, and therefore may be best viewed as a synergistic, rather than a competing, technology for improving guidance during soft-tissue surgeries.

DATA AVAILABILITY STATEMENT

The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials.