Multimodal fusion of DCE-MRI habitat–peritumoral radiomics and mammographic deep learning features for early recurrence prediction and risk stratification in invasive breast cancer

Background Invasive breast cancer exhibits marked biological heterogeneity, resulting in variable postoperative recurrence risks among patients with similar clinicopathological characteristics. This study aimed to develop a multimodal combined model integrating DCE-MRI-derived habitat–peritumoral radiomics, mammographic deep learning features, and clinicoradiological variables for early recurrence prediction and risk stratification. Methods A total of 419 patients with histologically confirmed invasive breast cancer who underwent surgery were retrospectively enrolled and randomly divided into training (n=293) and test (n=126) cohorts. Radiomics features were extracted from intratumoral habitat subregions and multi-scale peritumoral regions using preoperative DCE-MRI. Dual-view mammographic deep features were extracted using an ImageNet-pretrained ResNet50 model. After feature selection, individual radiomics, deep learning, and clinicoradiological models were developed, followed by construction of a multimodal Combined model using logistic regression. Model performance was assessed using ROC analysis, DeLong test, calibration curves, decision curve analysis, and risk stratification analysis. Results The 3-mm peritumoral region showed the optimal predictive performance among different expansion scales. Age, PR status, N3 stage, adjacent vessel sign, and non-mass-like morphology were identified as independent clinicoradiological predictors. The peritumoral–habitat radiomics model achieved AUCs of 0.876 and 0.864 in the training and test cohorts, respectively, demonstrating strong predictive performance. The Combined model showed further increases in AUC, reaching 0.922 and 0.893 in the two cohorts, respectively; however, its advantage over the PH model did not reach statistical significance in the test cohort. In addition, the Combined model achieved recurrence risk stratification in both cohorts. Conclusions The proposed multimodal fusion framework integrates complementary imaging and clinical information, providing a non-invasive and interpretable approach for early recurrence prediction and risk stratification in invasive breast cancer. However, the findings are based on internal validation conducted in a single center, and the model’s generalizability and clinical applicability remain to be established. Future research should include prospective, multi-center, external validation studies to further confirmed.

Authors

Institutions

Publication Details

Journal
Frontiers in Oncology
Published
2026-11-18
DOI
https://doi.org/10.3389/fonc.2026.1999806
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multimodal fusion of DCE-MRI habitat–peritumoral radiomics and mammographic deep learning features for early recurrence prediction and risk stratification in invasive breast cancer

Tong Su, Kaidi Long, Zubin Ouyang, Wenyi Bao
6 links
Frontiers in Oncology
Radiomics and Machine Learning in Medical Imaging
article

Multimodal fusion of DCE-MRI habitat–peritumoral radiomics and mammographic deep learning features for early recurrence prediction and risk stratification in invasive breast cancer

Tong Su, Kaidi Long, Zubin Ouyang, Wenyi Bao
article en

Abstract

Background Invasive breast cancer exhibits marked biological heterogeneity, resulting in variable postoperative recurrence risks among patients with similar clinicopathological characteristics. This study aimed to develop a multimodal combined model integrating DCE-MRI-derived habitat–peritumoral radiomics, mammographic deep learning features, and clinicoradiological variables for early recurrence prediction and risk stratification. Methods A total of 419 patients with histologically confirmed invasive breast cancer who underwent surgery were retrospectively enrolled and randomly divided into training (n=293) and test (n=126) cohorts. Radiomics features were extracted from intratumoral habitat subregions and multi-scale peritumoral regions using preoperative DCE-MRI. Dual-view mammographic deep features were extracted using an ImageNet-pretrained ResNet50 model. After feature selection, individual radiomics, deep learning, and clinicoradiological models were developed, followed by construction of a multimodal Combined model using logistic regression. Model performance was assessed using ROC analysis, DeLong test, calibration curves, decision curve analysis, and risk stratification analysis. Results The 3-mm peritumoral region showed the optimal predictive performance among different expansion scales. Age, PR status, N3 stage, adjacent vessel sign, and non-mass-like morphology were identified as independent clinicoradiological predictors. The peritumoral–habitat radiomics model achieved AUCs of 0.876 and 0.864 in the training and test cohorts, respectively, demonstrating strong predictive performance. The Combined model showed further increases in AUC, reaching 0.922 and 0.893 in the two cohorts, respectively; however, its advantage over the PH model did not reach statistical significance in the test cohort. In addition, the Combined model achieved recurrence risk stratification in both cohorts. Conclusions The proposed multimodal fusion framework integrates complementary imaging and clinical information, providing a non-invasive and interpretable approach for early recurrence prediction and risk stratification in invasive breast cancer. However, the findings are based on internal validation conducted in a single center, and the model’s generalizability and clinical applicability remain to be established. Future research should include prospective, multi-center, external validation studies to further confirmed.

Frontiers in OncologyVol. 16
Second Affiliated Hospital of Chongqing Medical University (CN), The Affiliated Yongchuan Hospital of Chongqing Medical University (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon