Related Experiment Video
Updated: Jul 2, 2026
06:42
Optimization of Breast Biopsy and Mastectomy Sample Collection Procedures for Biobanking, Personalized Medicine, and Research Applications
Published on: September 2, 2025
Ultrasound-Based AI in Predicting Hormone Receptor Status in Breast Cancer: Is "Digital Biopsy" Possible
Beyza Nur Kuzan1, Can Ilgin2, Murat Emeç3
1Department of Radiology, Kartal Dr. Lütfi Kırdar City Hospital, Istanbul, Türkiye.
Journal of Imaging Informatics in Medicine
|July 1, 2026
Summary
This study introduces a "digital biopsy" using machine learning and ultrasound radiomics to predict breast cancer hormone receptor status non-invasively. The approach shows promise for personalized treatment by analyzing image features instead of tissue samples.
Area of Science:
- Radiology and Medical Imaging
- Oncology
- Artificial Intelligence in Medicine
Background:
- Accurate breast cancer histopathology and molecular subtyping are crucial for prognosis and treatment.
- Traditional biopsies are invasive and time-consuming.
- Non-invasive methods for predicting tumor characteristics are needed.
Purpose of the Study:
- To evaluate the feasibility of a 'digital biopsy' using machine learning models.
- To predict hormone receptor status (Estrogen Receptor, Progesterone Receptor) from ultrasound radiomic features non-invasively.
- To assess the potential for predicting C-erbB2, Ki-67, histological grade, and molecular subtype.
Main Methods:
- Retrospective analysis of 353 breast tumors from 311 patients.
- Extraction of radiomic features from pre-biopsy ultrasound images.
- Development of machine learning models (XGBoost, ExtraTrees) using top 10 features to predict tumor characteristics.
Main Results:
- Estrogen Receptor (ER) prediction achieved accuracy between 0.79-0.83, with ExtraTrees showing an AUC of 0.70.
- Progesterone Receptor (PR) prediction accuracy was 0.73, with XGBoost AUC of 0.61.
- Promising predictive performance was observed for C-erbB2, Ki-67, histological grade, and molecular subtype.
Conclusions:
- Ultrasound radiomic features combined with machine learning can effectively identify patterns associated with hormone receptor status.
- This 'digital biopsy' approach offers a non-invasive alternative to tissue sampling, potentially reducing patient burden.
- Further validation in larger, prospective studies is required for clinical integration and personalized treatment planning.