Related Experiment Video
Updated: May 14, 2025
08:49
Exosomal miRNA Analysis in Non-small Cell Lung Cancer NSCLC Patients' Plasma Through qPCR: A Feasible Liquid Biopsy Tool
Published on: May 27, 2016
11.0K
A machine-learning approach for pancreatic neoplasia classification based on plasma extracellular vesicles
Ioanna Angelioudaki1, Angeliki Iosif2, Konstadina Kourou2
12nd Department of Surgery, Aretaieion Hospital, Medical School of Athens, National and Kapodistrian University of Athens, Athens, Greece.
Frontiers in Oncology
|May 12, 2025
Summary
This study developed a machine learning pipeline using Extracellular Vesicles (EVs) and clinical data to accurately detect pancreatic cancer. EV-based features offer improved diagnostic capacity beyond current biomarkers for early detection.
Area of Science:
- Oncology
- Biotechnology
- Machine Learning
Background:
- Pancreatic cancer (PC) is a lethal malignancy with limited early detection methods.
- Liquid biopsy, particularly Extracellular Vesicles (EVs), shows promise for non-invasive cancer diagnosis.
- EVs carry biomarkers reflecting their cell of origin, aiding in cancer detection.
Purpose of the Study:
- To develop a machine learning (ML) pipeline for predicting pancreatic tumors using clinical variables and EV-based features.
- To differentiate between exocrine/endocrine pancreatic neoplasms, benign lesions, and non-oncological patients.
- To assess the diagnostic value of EV subpopulations in plasma for pancreatic cancer detection.
Main Methods:
- Collected plasma samples (N=126) and clinical data pre-surgery.
- Characterized EVs using flow cytometry-immunostaining, analyzing size and biomarkers (CD45, CD63, EphA2).
- Applied various ML algorithms (Logistic Regression, Random Forest, SVM, XGBoost) for 3-class classification, evaluating with AUC-ROC and Shapley values.
Main Results:
- Hematological and biochemical features were identified as significant predictors.
- ML models utilizing plasma EV subpopulations achieved high accuracy (>0.90).
- Random Forest and XGBoost algorithms demonstrated superior performance, with accuracies of 0.96 +/- 0.03 and 0.93 +/- 0.04 for the two use cases, respectively.
Conclusions:
- The ML-driven pipeline integrating clinical and EV data accurately predicts pancreatic tumors.
- EV-based features offer added diagnostic value, surpassing traditional biomarkers like CEA and CA19.9.
- This approach supports the feasibility of early pancreatic cancer detection through enhanced risk estimation.
