Related Experiment Video
Updated: Jul 11, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Interpretable Deep Learning Radiomics for Differentiating Pleomorphic Adenoma and Warthin Tumor
Chuyuan Ma1, Yunxia Huang2, Ziheng Huang1
1The Sixth School of Clinical Medicine, the Affiliated Qingyuan Hospital (Qingyuan People's Hospital), Guangzhou Medical University, Qingyuan, P.R. China.
In Vivo (Athens, Greece)
|June 30, 2026
Summary
An interpretable machine learning model accurately differentiates pleomorphic adenoma (PA) from Warthin tumor (WT) using computed tomography (CT) scans. This AI approach integrates deep learning radiomics and clinical data for improved surgical planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate preoperative differentiation between pleomorphic adenoma (PA) and Warthin tumor (WT) is crucial for surgical strategy.
- Conventional computed tomography (CT) has limitations in distinguishing these salivary gland tumors.
Purpose of the Study:
- To develop and validate an interpretable machine learning (ML) framework for enhanced diagnostic precision.
- To integrate clinical data, radiomics, and deep learning features from CT for improved PA vs. WT differentiation.
Main Methods:
- Retrospective analysis of 171 patients (84 PA, 87 WT) with preoperative CT scans.
- Extraction of 1,561 radiomic and 2,048 deep learning features, followed by LASSO selection.
- Construction and validation of ML models (XGBoost) using a 70:30 split, evaluated by AUC and SHAP analysis.
Main Results:
- The combined ML model significantly outperformed individual clinical or radiomic models.
- The XGBoost classifier achieved a validation AUC of 0.961 (95%CI=0.916-1.000).
- SHAP analysis identified deep-learning score, age, and gender as key predictors.
Conclusions:
- An interpretable ML model integrating deep learning radiomics and clinical data offers robust accuracy in distinguishing PA from WT.
- The use of SHAP values provides transparent insights for clinicians, aiding personalized treatment planning.