Objective:
Tumor resection continues to be the primary modality of treatment in most head and neck cancers. Since surgeons typically rely on the macroscopic appearance of the tumor, negative margins are not guaranteed; close surgical margins (<5mm) are found in 15-30% of head and neck cancer cases.

Recent advances in fluorescence-guided surgical techniques have aided surgeons in better intra-operative delineation of tumor margins, thereby reducing the possibility of positive or close margins. The ‘Advanced Firefly™’ (Intuitive, Sunnyvale, CA) is one such imaging advancement available in transoral robotic surgery (TORS) to detect tumor-specific fluorescence in patients who have been administered preoperatively with an optical agent, such as panitumumab-IRdye800.

Here, we evaluate the accuracy of normalized fluorescence signal to delineate tumor from normal tissue in patients receiving panitumumab-IRdye800 prior to TORS.

Methods: Consenting patients (n=4) undergoing TORS for biopsy-confirmed oropharyngeal squamous cell carcinoma received 50mg panitumumab-IRDye800 48 hours prior to surgery. The patients underwent TORS with tumor resection as the standard of care. Fluorescence imaging was performed using ‘Advanced Firefly'. Images were processed using MATLAB™ and the brightest frames were selected. Various areas of interest from these frames were analyzed using ImageJ™. To achieve a standardized fluorescence ratio (SFR) for comparison, fluorescence counts from areas of interest were collected and divided by fluorescent counts acquired from the robot manipulator instrument within the same field of view. The areas of interest were then correlated with histology and an SFR threshold was identified that delineated tumor from normal tissue.

Results: With the available 2000 data points from one patient, pathology-confirmed tumor tissues produced 63.3 ± 8.7 counts which were significantly higher (p=0.02) than pathology-confirmed normal tissues, 2.1 ± 1.7 counts. The mean counts of the instrument were on average 49% lower than the tumor tissues and 15% higher than normal tissues within each frame analyzed. The dynamic ranges of tumor, normal, and instrument counts were 32.0-81.2, 0.0-8.3 and 0.1-82.0 counts, respectively. The mean SFR for tumor tissues (254.5 ± 11.5) were significantly higher (p=0.02) than normal tissues (3.1 ± 0.33). Receiver operator characteristic curves revealed an SFR threshold of 2.0, with a sensitivity of 93.7% and specificity of 7.8%, to predict the presence of cancer in the field of view.

Conclusion: This technique can be used for surgeries in the future to predict the likelihood of cancer to ensure more accurate and precise resection of the tumor, thus increasing the probability of negative margins.