Related Experiment Video
Updated: Aug 5, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Integrated Intratumoral and Peritumoral Ultrasound Radiomics Models for Breast Nodule Diagnosis Using Machine
Xinru Yang1, Jiaxin Zuo1, Cuiping Liu2
1Department of Ultrasound Medicine, Shanghai Ninth People's Hospital, Shanghai Jiao Tong University School of Medicine, 200011 Shanghai, China.
Summary
This study shows that combining ultrasound radiomics from inside and around breast nodules improves accuracy in distinguishing benign from malignant cases. This approach aids clinical decisions and reduces unnecessary biopsies for better breast cancer diagnosis.
Area of Science:
- Medical Imaging
- Radiology
- Oncology
Background:
- Differentiating benign and malignant breast nodules, especially BI-RADS 3-4, is challenging due to ultrasound subjectivity.
- Conventional ultrasound assessment has operator-dependent limitations.
Purpose of the Study:
- To assess the diagnostic value of intratumoral and peritumoral ultrasound radiomics features.
- To distinguish benign from malignant BI-RADS 3-4 breast nodules.
- To develop interpretable machine learning models for breast nodule classification.
Main Methods:
- Retrospective analysis of ultrasound images from 1231 female patients across two institutions.
- Extraction of radiomics features from intratumoral and peritumoral regions using PyRadiomics.
- Screening of features using t-test, Spearman correlation, and LASSO regression.
- Training and comparison of ten machine learning models, including SVM, RF, and XGBoost.
- Interpretation of feature importance using SHAP analysis.
Main Results:
- The support vector machine (SVM) model using a 3-pixel transitional zone (EI3_SVM) achieved the highest AUC (0.875) in the external test set, outperforming the tumor core model (AUC 0.787).
- Peritumoral radiomics models generally showed higher or comparable AUCs to intratumoral models.
- SHAP analysis revealed that texture and shape features were key drivers of high specificity.
Conclusions:
- Integrating intratumoral and peritumoral radiomics features significantly enhances diagnostic accuracy and objectivity for breast nodules.
- This radiomics approach can improve clinical decision-making, reduce unnecessary biopsies, and support early, precise breast cancer diagnosis.
