Related Experiment Video
Updated: Aug 14, 2026
06:57
Utilizing High Resolution Ultrasound to Monitor Tumor Onset and Growth in Genetically Engineered Pancreatic Cancer Models
Published on: April 7, 2018
Explainable Machine Learning for Detecting Pancreatic Cancer from Structured Endoscopic Ultrasound Data: A
Nunzio Zignani1, Marco Balzarini2, Gloria Lopiano3
1Department of Pathophysiology and Transplantation, University of Milan, 20122 Milan, Italy.
Journal of Clinical Medicine
|August 13, 2026
Summary
Machine learning models using endoscopic ultrasound data can accurately diagnose pancreatic cancer. Decision trees offer a good balance of interpretability and performance for clinical use.
Area of Science:
- Oncology
- Medical Imaging
- Machine Learning
Background:
- Machine learning (ML) in medicine requires transparent, clinically relevant models.
- Interpreting raw imaging data is challenging for ML adoption in gastrointestinal oncology.
- Structured Endoscopic Ultrasound (EUS) features are underutilized in pancreatic cancer predictive modeling.
Purpose of the Study:
- To evaluate ML models for diagnosing pancreatic ductal adenocarcinoma (PDAC) using EUS variables.
- To assess the performance and interpretability of these ML models.
Main Methods:
- Retrospective multicenter study (n=761) with data from 2015-2023.
- Development and evaluation of classifiers including decision trees, random forests, and naïve Bayes.
- Performance assessment via discriminative ability, calibration, and selective prediction.
Main Results:
- All models achieved high discriminative performance (AUC ≥ 0.90).
- Decision trees offered the best interpretability-accuracy balance (balanced accuracy = 0.87, sensitivity = 0.89).
- Models demonstrated robustness in calibration and selective prediction.
Conclusions:
- Interpretable and high-performing ML models are feasible for PDAC diagnosis.
- These models can be implemented in real-life endoscopic settings.
- Routine EUS data can be effectively leveraged for advanced diagnostics.