Early and Prediagnostic Detection of Pancreatic Cancer from Computed Tomography

TL;DR

This study presents ePAI, an automated, three-stage CT-based system for early PDAC detection and localization, trained on 1,598 abdominal CTs and augmented with synthetic lesions to boost small-tumor sensitivity. In internal and external multicenter tests, ePAI achieves high AUCs ($0.985$ internal; $0.971$ external) and strong sensitivity for small tumors, with precise localization to pancreatic regions. It demonstrates feasibility of prediagnostic detection, identifying 75/159 cases up to 36 months before clinical diagnosis with a median lead time of $347$ days, and significantly outperforms a panel of 30 radiologists in sensitivity. The findings support ePAI as an assistive tool to accelerate early PDAC detection and targeted workups, though prospective validation and integration with risk stratification are needed to optimize real-world adoption and reduce false positives.

Abstract

Pancreatic ductal adenocarcinoma (PDAC), one of the deadliest solid malignancies, is often detected at a late and inoperable stage. Retrospective reviews of prediagnostic CT scans, when conducted by expert radiologists aware that the patient later developed PDAC, frequently reveal lesions that were previously overlooked. To help detecting these lesions earlier, we developed an automated system named ePAI (early Pancreatic cancer detection with Artificial Intelligence). It was trained on data from 1,598 patients from a single medical center. In the internal test involving 1,009 patients, ePAI achieved an area under the receiver operating characteristic curve (AUC) of 0.939-0.999, a sensitivity of 95.3%, and a specificity of 98.7% for detecting small PDAC less than 2 cm in diameter, precisely localizing PDAC as small as 2 mm. In an external test involving 7,158 patients across 6 centers, ePAI achieved an AUC of 0.918-0.945, a sensitivity of 91.5%, and a specificity of 88.0%, precisely localizing PDAC as small as 5 mm. Importantly, ePAI detected PDACs on prediagnostic CT scans obtained 3 to 36 months before clinical diagnosis that had originally been overlooked by radiologists. It successfully detected and localized PDACs in 75 of 159 patients, with a median lead time of 347 days before clinical diagnosis. Our multi-reader study showed that ePAI significantly outperformed 30 board-certified radiologists by 50.3% (P < 0.05) in sensitivity while maintaining a comparable specificity of 95.4% in detecting PDACs early and prediagnostic. These findings suggest its potential of ePAI as an assistive tool to improve early detection of pancreatic cancer.

Figures (4)

• Figure 1: Overview and performance of the ePAI system.A. The ePAI system detects pancreatic ductal adenocarcinoma (PDAC) through a three-stage cascade: Stage 1 segments pancreatic and surrounding anatomy, Stage 2 localizes all potential lesions, and Stage 3 classifies each lesion as PDAC, non-PDAC, or normal. B. Distribution of number of detections in prediagnostic CT scans, showing that ePAI identified PDAC a median of 347 days before the first clinical diagnosis by radiologists.
• Figure 2: Internal and external validation of early PDAC detection with ePAI.(A) Sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) of ePAI for detecting all-size PDAC in the internal test cohort and six external regional multicenter test cohorts. (B) Receiver operating characteristic (ROC) curves of ePAI for detecting all-size PDAC across internal and external cohorts, highlighting its generalizability across centers. (C) Sensitivity of ePAI stratified by PDAC lesion size ($\leq 2$ cm vs. $> 2$ cm). (D) Sensitivity stratified by PDAC T-stage (T1–T4), showing strong performance even in the earliest stages. Localization performance in internal (E) and external cohorts (F), evaluated by localization accuracy between the predicted lesion location and radiologists’ voxel-wise annotations or report-based localization, ePAI achieved 94.6% in internal test cohort while achieved 88.7% in external test cohort.
• Figure 3: Earlier PDAC detection in prediagnostic CT scans. Representative paired prediagnostic (left) and diagnostic (right) contrast-enhanced CT images from six patients in the external multicenter prediagnostic cohorts (3–36 months before first clinical diagnosis). Red contours indicate ePAI predictions or radiologist-confirmed tumor regions on the diagnostic scans. (A) Patients #1–#3 show direct prediagnostic detection: ePAI marks small, subtle abnormalities on the prediagnostic scans (small red contours), and the diagnostic scans later show an obvious PDAC at the same location (red contours), detected by both ePAI and radiologists. (B) Patients #4–#6 show indirect prediagnostic cues: ePAI highlights secondary changes on the prediagnostic scans, most notably pancreatic duct dilatation or cutoff (red 3D rendering), even when no clear mass is visible to ePAI or radiologists. The paired diagnostic scans later show a clear PDAC (red contour) detected by both ePAI and radiologists. Together, these examples illustrate two common prediagnostic patterns---early subtle lesions and secondary ductal changes before an obvious mass---and potentially support the use of ePAI for earlier PDAC review.
• Figure 4: Multi-reader study. (A) Comparison between ePAI and 30 readers with different expertise, evaluated on prediagnostic, diagnostic, and normal CT scans. ePAI shows substantially higher sensitivity in both prediagnostic and diagnostic settings while maintaining high specificity on normal controls. (B) Balanced accuracy of individual readers (pancreatic imaging specialists, general radiologists, and radiology residents) compared with ePAI and themselves with assistance of ePAI. (C) Sensitivity, specificity of ePAI, readers, and reader-plus- ePAI combinations for prediagnostic, diagnostic, and normal cohorts.