ENDOSWIR clinical proof of concept: a reflectance-based multispectral imaging device for head and neck cancer detection.

Achieving clear surgical margins during head and neck squamous cell carcinoma (HNSCC) procedures is of paramount significance, as it plays a crucial role in mitigating the probability of recurrence and enhancing patient outcomes. The differentiation of tumors from healthy tissue remains a significant challenge, particularly in cases of infiltrative tumors. The integration of multispectral short-wave infrared (SWIR) imaging with machine learning algorithms may enhance the efficacy of tumor detection, particularly in the specific areas of resection margins. In this prospective monocentric study (2022), 10 patients with oral or oropharyngeal squamous cell carcinoma were included. The ENDOSWIR device, a multispectral SWIR imaging system, was utilized in the analysis of freshly resected ex vivo specimens. Reflectance at six wavelengths (975-1575 nm) was recorded. A support vector machine (SVM) algorithm was employed to differentiate between tumoral and healthy tissues. This classification was achieved through the analysis of reflectance absolute values and slope features. Subsequently, a binary mask was applied to the images, utilizing pixel-wise SVM scores. Histopathology served as the gold standard for the diagnosis of these conditions. The primary outcomes encompassed the sensitivity, specificity, and predictive values of ENDOSWIR. Secondary outcomes assessed processing time in comparison to frozen section analysis. Across a total of 494 regions of interest, the SVM algorithm demonstrated a 95% sensitivity and 96% specificity, exhibiting positive and negative predictive values of 95%. On independent test datasets, sensitivity and specificity levels were recorded at 80% and 88%, respectively. The acquisition and analysis of ENDOSWIR samples required a mean time of 3-5 minutes per sample, a significant reduction compared to the approximately 23 minutes required for frozen section analysis. ENDOSWIR exhibited high precision and rapidity in detecting ex vivo tumor tissue, suggesting its potential as a rapid, effective tool to improve surgical precision in HNSCC resections and assist pathologists. Subsequent optimization is necessary for intraoperative utilization.