A Multimodal Fusion Model of Radiomics and Deep Learning Integrating the Tumor Microenvironment Accurately Predicts...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Predicting pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer is crucial but challenging due to tumor heterogeneity and imaging limitations.
• Existing radiomics and deep learning (DL) models often overlook the peritumoral microenvironment, a key factor in treatment response.

Purpose of the Study:

• To develop and validate a multimodal model integrating intratumoral radiomics, peritumoral features, and DL for enhanced pCR prediction in breast cancer.
• To assess the model's performance against single-modality approaches and evaluate its clinical utility.

Main Methods:

• A multimodal model was created using pre-NAC MRI data, combining intratumoral radiomics, peritumoral features (9-mm expansion), and DL patterns.
• The model was trained and validated on multicenter cohorts (I-SPY2 trial, n=929; independent cohort, n=95) using Lasso regression and bidirectional selection for feature optimization.
• Logistic regression was identified as the optimal machine learning algorithm for the integrated model (Intra-Peri-DL).

Main Results:

• The integrated Intra-Peri-DL model achieved high predictive performance with an AUC of 0.888 (internal) and 0.890 (external validation).
• The model demonstrated statistically superior performance compared to single-modality radiomics or DL models (P<0.05).
• High sensitivity (>0.91) was observed in both cohorts, indicating strong potential for clinical application in treatment decision-making.

Conclusions:

• The developed multimodal model effectively synergizes radiomics and DL to capture tumor-microenvironment interactions, significantly improving pCR prediction accuracy.
• This approach offers a promising, clinically actionable tool for personalized neoadjuvant chemotherapy strategies in breast cancer.
• The framework bridges imaging data with biological insights, advancing precision oncology for breast cancer patients.