Related Experiment Video
Updated: Oct 6, 2025
06:21
Ultrasound-Guided Orthotopic Implantation of Murine Pancreatic Ductal Adenocarcinoma
Published on: November 19, 2019
11.5K
Fully Automatic Deep Learning Framework for Pancreatic Ductal Adenocarcinoma Detection on Computed Tomography
Natália Alves1, Megan Schuurmans1, Geke Litjens2
1Diagnostic Image Analysis Group, Department of Medical Imaging, Radboud University Medical Center, 6500 HB Nijmegen, The Netherlands.
Cancers
|January 21, 2022
Summary
Deep learning models can now detect small pancreatic ductal adenocarcinoma (PDAC) lesions on CT scans. Integrating surrounding anatomy information significantly improves the accuracy of these artificial intelligence tools for early PDAC diagnosis.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Early detection of pancreatic ductal adenocarcinoma (PDAC) is crucial for improving patient prognosis.
- Diagnosing small or poorly defined PDAC lesions on contrast-enhanced computed tomography (CE-CT) scans remains a significant challenge.
- Existing deep learning models often struggle with the accurate identification of small (<2 cm) PDAC lesions.
Purpose of the Study:
- To develop an automatic deep learning framework for enhanced detection and segmentation of PDAC, with a specific focus on small lesions.
- To investigate the impact of incorporating surrounding anatomical information into deep learning models for PDAC detection.
- To compare the performance of different deep learning architectures in identifying PDAC on CE-CT scans.
Main Methods:
- Trained three nnU-Net models using CE-CT scans from 119 PDAC patients and 123 controls.
- nnUnet_T: Focused on automatic lesion detection and segmentation.
- nnUnet_TP: Included segmentation of the pancreas and tumor.
- nnUnet_MS: Incorporated segmentation of the pancreas, tumor, and surrounding anatomical structures.
- Evaluated model performance on an external, publicly available test dataset.
Main Results:
- The nnUnet_MS model, which integrated surrounding anatomy, demonstrated superior performance.
- Achieved an area under the receiver operating characteristic curve (AUC) of 0.91 for the entire test set.
- Specifically for tumors <2 cm, the nnUnet_MS achieved an AUC of 0.88, indicating effectiveness in detecting small lesions.
Conclusions:
- State-of-the-art deep learning models can effectively detect small pancreatic ductal adenocarcinoma lesions.
- Integrating information about surrounding anatomy significantly enhances the performance of deep learning models for PDAC detection.
- This approach holds promise for improving early diagnosis and patient outcomes in PDAC.
