Related Experiment Video
Updated: Jun 26, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Radiomics-Guided Multi-Sequence Learning for Pathological Complete Response Prediction from Breast MRI with Missing
Xinyuan Xiang1, Wenyu Yin1, Jiayue Li2
1School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China.
Journal of Imaging
|June 25, 2026
Summary
This study introduces a radiomics-guided framework to predict pathological complete response (pCR) in breast cancer using multi-sequence MRI. The novel approach enhances prediction accuracy compared to standard methods.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Pathological complete response (pCR) is a critical endpoint for neoadjuvant chemotherapy (NACT) in breast cancer treatment.
- Multi-sequence breast MRI aids in pCR prediction, but standard MRI protocols may not always include all necessary sequences (e.g., T1-weighted, T2-weighted).
- Current models often combine radiomic and deep features via simple concatenation, inadequately capturing interactions between handcrafted and learned representations.
Purpose of the Study:
- To develop and evaluate a novel radiomics-guided framework for predicting pCR from multi-sequence breast MRI.
- To improve upon existing methods by better integrating radiomic features with deep learning representations derived from various MRI sequences.
- To assess the feasibility and performance of the proposed framework in a clinical cohort.
Main Methods:
- Development of a multi-branch 2.5D encoder to extract sequence-specific features from multi-sequence breast MRI.
- Implementation of radiomics-guided channel recalibration to enhance feature representation.
- Utilization of masked token fusion to effectively aggregate features from available MRI sequences.
- Evaluation using a 5-fold cross-validation on 157 patients from the I-SPY1 Trial cohort.
Main Results:
- The full radiomics-guided framework achieved 78.4% accuracy and 0.809 Area Under the Curve (AUC).
- This performance surpassed the strongest baseline model (channel-concatenation), which yielded 75.8% accuracy and 0.788 AUC.
- The study demonstrated the feasibility of radiomics-guided multi-sequence learning for pCR prediction in this cohort.
Conclusions:
- The developed radiomics-guided framework shows promise for improving pCR prediction in breast cancer using multi-sequence MRI.
- This approach offers a more sophisticated method for integrating diverse imaging features compared to simple concatenation.
- Further external validation is necessary to confirm clinical applicability and generalizability of the findings.
