Validation of full-field optical coherence tomography in distinguishing malignant and benign tissue in resected pancreatic cancer specimens

Abstract

Background: Pancreatic cancer is the fourth leading cause of cancer-related mortality in the United States. The minority of patients can undergo curative-intended surgical therapy due to progressive disease stage at time of diagnosis. Nonetheless, tumor involvement of surgical margins is seen in up to 70% of resections, being a strong negative prognostic factor. Real-time intraoperative imaging modalities may aid surgeons to obtain tumor-free resection margins. Full-field optical coherence tomography (FF-OCT) is a promising diagnostic tool using high-resolution white-light interference microscopy without tissue processing. Therefore, we composed an atlas of FF-OCT images of malignant and benign pancreatic tissue, and investigated the accuracy with which the pathologists could distinguish these.

Materials and methods: One hundred FF-OCT images were collected from specimens of 29 patients who underwent pancreatic resection for various indications between 2014 and 2016. One experienced gastrointestinal pathologist and one pathologist in training scored independently the FF-OCT images as malignant or benign blinded to the final pathology conclusion. Results were compared to those obtained with standard hematoxylin and eosin (H&E) slides.

Results: Overall, combined test characteristics of both pathologists showed a sensitivity of 72%, specificity of 74%, positive predictive value of 69%, negative predictive value of 79% and an overall accuracy of 73%. In the subset of pancreatic ductal adenocarcinoma patients, 97% of the FF-OCT images (n = 35) were interpreted as tumor by at least one pathologist. Moreover, normal pancreatic tissue was recognised in all cases by at least one pathologist. However, atrophy and fibrosis, serous cystadenoma and neuroendocrine tumors were more often wrongly scored, in 63%, 100% and 25% respectively.

Conclusion: FF-OCT could distinguish normal pancreatic tissue from pathologic pancreatic tissue in both processed as non-processed specimens using architectural features. The accuracy in pancreatic ductal adenocarcinoma is promising and warrants further evaluation using improved assessment criteria.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Fig 1. Example FF-OCT image of a well-differentiated pancreatic ductal adenocarcinoma as shown to the pathologist for assessment.

The whole FF-OCT images were shown to the pathologists, but they were asked to only assess the selected regions of interest (two per FF-OCT image). Both the original (A,C,E) and the inverse (B,D,F) FF-OCT images are shown. Scale bars, 5 mm (A,B), 2 mm (C,D), and 1 mm (E,F).

Fig 2. Examples of FF-OCT images of normal pancreatic tissue and corresponding histology.

Structures that are easily identified on FF-OCT images include normal pancreatic parenchyma (A-B), vessels (C-D), and a large pancreatic duct (E-F). Harder to recognize are nerve bundles (G-H). Scale bars, 250 μm (A-B, G-H), and 150 μm (C-F). Inset shows an islet of Langerhans (A-B) at 2.5 times higher magnification.

Fig 3. Example of benign pancreatic tissue FF-OCT images and corresponding histology.

Fibrotic pancreatic tissue after neoadjuvant therapy (A-B), pancreatitis (C-D), and a serous cystadenoma (E-F). Scale bars all 500 μm.

Fig 4. Example FF-OCT image of a well differentiated pancreatic ductal adenocarcinoma.

A shows a n overview. B shows a magnified view of stromal disorganization, and C shows nests of tumor cells. Scale bars, 2 mm (A), 1 mm (B), and 500 μm (C).

Fig 5. Accuracy and interobserver variability.

Pathologist 1 is the experienced pathologist, pathologist 2 is the pathologist in training.

Fig 6. Proposed decision tree to evaluate pancreatic FF-OCT images.

Scalebars are all 500 μm. *The FF-OCT image and corresponding histology image only depict lymph node invasion.