Related Experiment Video
Updated: May 3, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.7K
Distinction Between Benign and Borderline/Malignant Phyllodes Tumor in Breast Mammography and Ultrasound Based on
Xiaohui Su1, Chao Li2, Jingjing Chen1
1Department of Breast Imaging, The Affiliated Hospital of Qingdao University, Qingdao, Shandong, China (X.S., J.C., C.C., T.B., L.L., N.S.).
Academic Radiology
|February 21, 2026
Summary
Radiomics analysis of mammography and ultrasound images effectively distinguishes benign from borderline/malignant phyllodes tumors (PTs). This advanced imaging technique aids in preoperative assessment and surgical planning for PTs, improving patient care.
Area of Science:
- Medical Imaging
- Oncology
- Radiology
Background:
- Phyllodes tumors (PTs) are rare breast neoplasms with unpredictable behavior.
- Accurate preoperative differentiation between benign and borderline/malignant PTs is crucial for appropriate management.
- Existing imaging modalities have limitations in definitively distinguishing PT subtypes.
Purpose of the Study:
- To evaluate the efficacy of radiomics applied to mammography (MG) and ultrasound (US) for differentiating benign from borderline/malignant phyllodes tumors (PTs).
- To compare the diagnostic performance of radiomics models against clinical and conventional imaging features.
Main Methods:
- Retrospective analysis of 362 female patients with PTs (220 benign, 142 borderline/malignant).
- Extraction of radiomics features from MG (CC, MLO views) and US images, alongside clinical data.
- Development and comparison of machine learning classifiers (LR, SVM, RF, ExtraTrees, XGBoost, LightGBM) and a combined nomogram model.
- Evaluation using Area Under the Curve (AUC), sensitivity, specificity, accuracy, PPV, and NPV.
Main Results:
- Borderline/malignant PTs were associated with older age and larger tumor size compared to benign PTs.
- Significant differences observed in MG (indistinct margins, heterogeneous density) and US (cystic changes, noncircumscribed margins) features.
- Radiomics model achieved an AUC of 1.0 in the training cohort, outperforming other models.
- In the validation cohort, the nomogram model showed the highest AUC (0.791), followed by MLO (0.777) and radiomics (0.766).
Conclusions:
- Radiomics and nomogram models show significant potential in differentiating benign and borderline/malignant PTs.
- These models can aid in guiding treatment strategies and improving preoperative assessment.
- A combined approach using core needle biopsy and the nomogram model is recommended for optimal decision support in PT management.
