Hyperspectral imaging based on compressive sensing to determine cancer margins in human pancreatic tissue ex vivo

Abstract

Objective: Pancreatic adenocarcinoma (PA) is the fourth-leading cause of cancer-related mortality in the United States, with a 5-year survival rate <10%. Surgical resection is the most widely-utilized treatment for pancreatic adenocarcinoma. Locally advanced lesions frequently invade vasculature and surrounding tissue, making R0 resections challenging. We sought to develop a single-pixel, hyperspectral imaging system based on compressive sensing for tumor margin detection. Methods: Freshly-resected pancreatic tissue from 10 patients with a diagnosis of presumptive PA were analyzed. A spectral imaging system comprising of a nitrogen laser (autofluorescence) and halogen lamp (reflectance) were used to generate spectra from tumor, non-tumor and tumor/non-tumor borders. Changes in autofluorescence and reflectance spectra were used to delineate tissue from tumor and non-tumor pancreas, and these data were compared to H&E stained sections following blinded scoring by a board-certified pathologist. Results: Our data indicated significant changes in both reflectance and autofluorescent spectra in non-tumor and tumor tissue from PA resections. Reflectance spectra was elevated at 450 nm in tumor tissue and decreased at 550 nm compared to non-tumor tissue. Autofluorescent spectra were elevated as a result of increased collagen (400 nm) and nicotinamide adenine dinucleotide phosphate (NAD(P)H) (475 nm) in tumor tissue compared to non-tumor tissue. A reflectance image formed from the ratio at 450 nm and 500 nm was sensitive to malignant tissue and showed excellent correlation to margins as determined following pathological analysis. Conclusion: Single-pixel hyperspectral imaging represents a novel technology to map tumor and non-tumor margins with high confidence.