Introduction
Endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB) is the standard for sampling solid pancreatic lesions, with up to 95% diagnostic accuracy. Ex vivo fluorescence confocal laser microscopy with the virtual on-site evaluation (VOSE) delivers near real-time microscopic images of a fresh unfixed biopsy and could shorten diagnostic pathway. We prospectively assessed histologic adequacy and diagnostic agreement of VOSE against conventional histology in patients undergoing EUS-FNB for a pancreatic mass
Aims & Methods
In this single center prospective study consecutive patients with a solid pancreatic lesion performing EUS-FNB were included. The samples were assessed by VOSE scored as benign or malignant, with a descriptive specific provisional diagnosis. The interobserver agreement between the new method and standard histological analysis were performed blindly between four pathologists. The routine histology was performed on the specimen analyzed by VOSE and on a separate EUS-FNB pass. The final diagnosis was based on EUS-FNB histology results and at least 6 months follow-up.
We calculated diagnostic yield parameters with 95% Wilson confidence intervals, likelihood ratios and Cohen's kappa. Descriptive labels were reviewed qualitatively to characterise misclassifications.
Results
Eighty-two patients were included (mean age 64.9 ± 11.8 years, range 30–88). Lesions were located in the head (n=45), uncinate (7), body (18) and tail (12). Mean size of the lesions was 25 ± 12 mm. Histology yielded 58 malignancies (adenocarcinoma 46, neuroendocrine tumour 5, metastases 3, lymphoma 2, IPMN 1, schwannoma 1) and 24 benign lesions (chronic pancreatitis 21, serous cystadenoma 1, IgG4-related pancreatitis 1, intrapancreatic spleen 1).
The paired 2×2 table was TP 57, FP 4, TN 20, FN 1. VOSE showed sensitivity 98.3% (95% CI 90.9–99.7), specificity 83.3% (64.1–93.3), PPV 93.4% (84.3–97.4), NPV 95.2% (77.3–99.2) and overall accuracy 93.9% (86.5–97.4). Agreement with histology was almost perfect (kappa 0.85, 95% CI 0.72–0.98), with no directional bias (McNemar exact p=0.38).
All four false positives were fragments interpreted on VOSE as neuroendocrine tumour (NET); the reference histology was chronic pancreatitis (n=3) and serous cystadenoma (n=1). Overall, VOSE applied a NET descriptor to nine fragments, but only three corresponded to a histological NET (descriptive PPV 33%); two further NETs described on Vivascope were correctly classified as malignant, but turned out to be adenocarcinomas in final histology. The single false negative was a lesion read on VOSE as chronic pancreatitis that proved to be ductal adenocarcinoma at histology.
At 6-month follow-up, 3 cases concordantly read as chronic pancreatitis by both VOSE and histology were re-classified as malignant (2 adenocarcinomas, 1 lymphoma), consistent with sampling errors rather than interpretive errors. Against this composite reference, the overall diagnostic performance of EUS-FNB was: sensitivity 95.1% (95% CI 86.5–98.3), specificity 100% (84.5–100), PPV 100%, NPV 87.5% (69.0–95.7) and accuracy 96.3% (89.8–98.7).
Conclusion
In this single-fragment EUS-FNB cohort, ex vivo Vivascope imaging showed excellent concordance with conventional histology for the benign/malignant call (accuracy 94%, kappa 0.85), with very high sensitivity and NPV supporting its use as a rapid adjunct during the procedure. Specificity was limited by over-recognition of a NET-like pattern, which should be interpreted cautiously and confirmed on standard histology before driving management decisions.