Radiomic Features at CT Can Distinguish Pancreatic Cancer from Noncancerous Pancreas

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Pancreatic ductal adenocarcinoma (PDAC) is a challenging diagnosis.
• CT imaging offers potential for non-invasive characterization.
• Radiomics extracts quantitative features from medical images.

Purpose of the Study:

• To identify CT radiomic features distinguishing PDAC.
• To develop a machine learning model for PDAC detection.
• To assess the model's ability to differentiate PDAC from normal pancreas.

Main Methods:

• Retrospective analysis of contrast-enhanced CT images from Taiwanese and U.S. cohorts.
• Image processing into patches for XGBoost model training.
• Patch-based classification to determine patient-level PDAC status.
• Development of local and generalized models for classification.

Main Results:

• PDAC showed lower intensity and higher heterogeneity radiomic features compared to normal pancreas.
• The generalized model achieved high performance: sensitivity 94.7%, specificity 95.4%, accuracy 95.0% on Taiwanese test data.
• Performance on U.S. test data: sensitivity 80.6%, specificity 100%, accuracy 86.5%.

Conclusions:

• Radiomic analysis with machine learning accurately detects PDAC on CT.
• The developed models can effectively identify patients with PDAC.
• This approach shows promise for computer-aided diagnosis of pancreatic cancer.