Endocrine Abstracts

Background: Pancreatic malignancies, including pancreatic neuroendocrine tumours (PNETs), present major diagnostic challenges. The gold standard for diagnosis is endoscopic ultrasound-guided fine-needle aspiration or biopsy (FNA/B), assessed by a cytopathologist. However, workforce shortages have prompted interest in computer-assisted diagnosis (CAD) to ease these pressures and improve diagnostic efficiency. Despite this, the collective diagnostic impact of CAD in pancreatic cytology has not been quantified. We aimed to evaluate the accuracy of CAD in this setting.

Methods: We conducted a systematic review and meta-analysis; five databases were searched for studies published between January 2010 and May 2025 applying any CAD technique to pancreatic FNA/B cytology. Two reviewers independently screened records, with risk of bias assessed using QUADAS-2. Random-effects bivariate models generated pooled sensitivity, specificity and summary receiver-operating-characteristic (SROC) curves. Studies were analysed separately depending on whether the CAD tool evaluated multiple cytopathological images per case (multi-image level) or a single image from each case (single-image level).

Results: Ten studies met eligibility criteria and provided quantitative data. At the multi-image level, pooled sensitivity and specificity were 91% [95% confidence interval: 86–94] and 92% [87–96], respectively. At the single-image level, pooled sensitivity and specificity were 85% [69–93] and 91% [73–97], respectively. The SROC area under the curve was 0.945. Heterogeneity was high (I² = 65–93%), driven by retrospective designs and variable reference standards.

Conclusion: CAD tools achieved near-expert diagnostic accuracy and could reduce indeterminate reports and staffing demands. In endocrine oncology, they may enhance diagnostic precision for pancreatic cancers, such as PNETs, enabling earlier hormonal evaluation and surgical planning. However, selection bias, single-centre training and inconsistent thresholds limit generalisability. Prospective multi-centre validation with whole-slide workflows and consistent reporting is warranted, and integration with telepathology may expand access in low-resource regions.