Predicting Neoadjuvant Chemotherapy Response in Breast Cancer Using Multiparametric MRI-Based Radiomics and Machine
Mohammadreza Elhaie1, Abolfazl Koozari2, Marzieh Sheikh3
1Department of Medical Physics, School of Medicine, Isfahan University of Medical Sciences, Isfahan, Iran.
Indian Journal of Surgical Oncology
|August 6, 2026
Summary
Multiparametric MRI (mpMRI)-based radiomics and machine learning (ML) show promise for predicting neoadjuvant chemotherapy response in breast cancer. Further standardized studies are needed for clinical application.
Area of Science:
- Oncology
- Radiology
- Data Science
Background:
- Breast cancer treatment response prediction is challenging due to tumor heterogeneity.
- Neoadjuvant chemotherapy (NAC) response is critical for personalized treatment strategies.
- Multiparametric magnetic resonance imaging (mpMRI)-based radiomics and machine learning (ML) offer non-invasive prediction methods.
Purpose of the Study:
- To systematically review and meta-analyze the predictive performance of mpMRI-based radiomics and ML for NAC response in breast cancer.
- To identify key radiomic features and ML algorithms used in prediction.
- To assess methodological limitations and future research directions.
Main Methods:
- Systematic review and meta-analysis following PRISMA guidelines, registered on PROSPERO.
- Searched major scientific databases (PubMed, Embase, etc.) for relevant studies.
- Extracted data on study characteristics, mpMRI sequences, radiomic features, ML algorithms, and predictive performance (AUC, sensitivity, specificity).
- Assessed study quality using QUADAS-2 and Radiomics Quality Score.
- Performed random-effects meta-analysis to pool AUC for pathological complete response (pCR) prediction.
Main Results:
- Included eight studies (22-328 patients), utilizing DCE, DWI, and T2-weighted mpMRI sequences.
- Common radiomic features included texture (entropy, kurtosis), ADC, and morphological parameters.
- ML algorithms ranged from logistic regression to deep neural networks, with reported AUCs from 0.609 to 0.94.
- The pooled AUC for pCR prediction was 0.803 (95% CI: 0.682-0.925).
- Identified limitations: small sample sizes, lack of external validation, and variable imaging protocols.
Conclusions:
- mpMRI-based radiomics combined with ML demonstrates significant potential for predicting breast cancer NAC response.
- Methodological inconsistencies and limitations highlight the need for standardized, multicenter research.
- Further validation is required before widespread clinical translation.
