Related Experiment Video
Updated: Jun 29, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.8K
An ultrasound-based ensemble machine learning model for the preoperative classification of pleomorphic adenoma and
Yanping He1, Bowen Zheng2, Weiwei Peng1
1Department of Medical Ultrasonics, The First People's Hospital of Foshan, No. 81, Lingnan Avenue North, Foshan, 528000, China.
European Radiology
|April 3, 2024
Summary
A new ultrasound-based ensemble machine learning (USEML) model accurately differentiates pleomorphic adenomas (PMA) and Warthin tumors (WT) in the parotid gland. This noninvasive model outperforms physicians, offering potential for improved diagnostic strategies.
Area of Science:
- Oncology
- Medical Imaging
- Machine Learning
Background:
- Accurate preoperative classification of parotid gland tumors, specifically pleomorphic adenoma (PMA) and Warthin tumor (WT), is crucial for guiding therapeutic strategies.
- Current diagnostic methods may involve invasive procedures, highlighting the need for noninvasive and reliable alternatives.
Purpose of the Study:
- To develop and validate a noninvasive ultrasound-based ensemble machine learning (USEML) model for differentiating PMA from WT in the parotid gland.
- To compare the diagnostic performance of the USEML model against traditional ultrasound (US) and clinical models, as well as experienced physicians.
Main Methods:
- A cohort of 203 patients with histologically confirmed PMA or WT was analyzed.
- Clinical, ultrasound (US), and radiomic features were extracted to build distinct machine learning models.
- The diagnostic performance was evaluated using receiver operating characteristic (ROC) curves and validated internally and externally, with SHAP values used for model interpretability.
Main Results:
- The USEML model demonstrated superior diagnostic performance with the highest Area Under the Curve (AUC) of 0.891, outperforming US (0.847) and clinical (0.814) models.
- The USEML model significantly outperformed experienced physicians in both internal and external validation cohorts (p < 0.05).
- Key performance metrics for the USEML model included high sensitivity (89.3%) and specificity (87.5%).
Conclusions:
- The developed USEML model effectively distinguishes between PMA and WT using a combination of clinical, ultrasound, and radiomic features.
- The model's superior performance compared to physicians suggests its potential as a valuable tool in clinical settings for preoperative parotid tumor diagnosis.
- This noninvasive approach holds promise for improving patient management by enabling more accurate preoperative classification.
