PancreaX: Artificial intelligence–driven diagnosis of pancreatic cancer.

Abstract

e16465 Background: Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies, with poor survival rates largely due to late-stage diagnosis. Distinguishing PDAC from chronic pancreatitis (CP) remains a major diagnostic challenge, as both conditions exhibit overlapping histopathological features. Traditional diagnostic methods, including histopathology, remain the gold standard but often yield inconclusive results due to the fragmented nature of biopsy samples and the subtlety of distinguishing features. This study explores the integration of artificial intelligence (AI)-driven computer vision techniques to enhance histologic assessments and improve the accuracy of PDAC diagnosis. By leveraging computational pathology, AI models can aid pathologists in identifying critical features, reducing diagnostic uncertainty, and facilitating earlier intervention. Methods: The study utilized a dataset of 518 whole-slide images (WSIs) of shark core biopsies from patients diagnosed with PDAC (n = 373) or CP (n = 145). A subset comprising 24 PDAC and 15 CP cases was reserved for testing. A board-certified gastrointestinal pathologist annotated 259 WSIs, and 256×256 pixel patches were extracted from these annotated regions, generating a total of 489,556 patches. The dataset was divided into 70% for training and 30% for validation. The model was developed and trained using ORCA™ (SpatialX Diagnostics, Inc., USA). Additionally, an independent validation cohort of 20 PDAC cases from TCGA was used to evaluate generalizability. Results: PancreaX demonstrated outstanding performance, achieving a sensitivity of 99% and specificity of 98%. The model accurately identified all PDAC cases in test and TCGA data. Validation on the independent cohort yielded consistent results, with a sensitivity of 96% without fine-tuning, underscoring the model’s robustness and ability to generalize across diverse datasets. Conclusions: PancreaX represents a clinically accessible AI-driven diagnostic tool that significantly enhances the accuracy and efficiency of pancreatic cancer diagnoses. By guiding pathologists to critical regions and prioritise cases, PancreaX facilitates faster therapeutic decisions.