Large-scale pancreatic cancer detection via non-contrast CT and deep learning
Problem
Pancreatic ductal adenocarcinoma is typically detected late and inoperable, and identifying it on non-contrast CT was long considered impossible.
Why it matters
PDAC is described as the most deadly solid malignancy, and early detection is associated with prolonged survival.
What the AI did
A deep learning approach (PANDA) detects and classifies pancreatic lesions on non-contrast CT, trained on 3,208 patients from a single center and validated across multiple centers.
Results
AUC of 0.986-0.996 for lesion detection in multicenter validation; trained on 3,208 patients.
Impact
Demonstrates that PDAC detection on non-contrast CT is feasible, potentially enabling opportunistic screening.
Record
| Venue | Nature Medicine |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Kai Cao, Yingda Xia, Jiawen Yao, Xu Han, Lukáš Lambert, Tingting Zhang, Wei Tang, Gang Jin et al. |
| Citations | 354 (OpenAlex, at import time) |
| References | 62 |
| Model or method | PANDA (deep learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41591-023-02640-w |
| Open access | hybrid |
| Keywords | pancreatic cancer, deep learning, non-contrast CT, PANDA, medical imaging, screening |
Importance rating
major — millions of lives, an industry or a field
Abstract states PDAC is the most deadly solid malignancy, typically detected late/inoperable, and early detection prolongs survival; enables large-scale screening via routine non-contrast CT.