Related Experiment Video
Updated: Sep 20, 2025
10:37
A Multimodal Imaging Framework to Advance Phenotyping of Living Label-free Breast Cancer Cells
Published on: August 22, 2025
327
An Explainable Multimodal Artificial Intelligence Model Integrating Histopathological Microenvironment and EHR
Zijian Yang1, Changyuan Guo2, Jiayi Li3,4,5
1Institute of Genomic Medicine, School of Biomedical Engineering, Wenzhou Medical University, Wenzhou, 325027, China.
Summary
Multimodal Artificial Intelligence Germline Genetic Testing (MAIGGT) accurately prescreens high-risk breast cancers for BRCA1/2 mutations by integrating pathology images and clinical data. This AI approach enhances genetic testing accessibility for hereditary breast cancer.
Area of Science:
- Oncology
- Genetics
- Artificial Intelligence
- Digital Pathology
Background:
- Germline pathogenic variants in BRCA1/2 genes significantly increase breast cancer risk.
- Personalized management of high-risk breast cancers relies on genetic testing for targeted therapies and family screening.
- Current genetic testing methods can be costly and inaccessible for widespread prescreening.
Purpose of the Study:
- To develop and validate a deep learning framework, MAIGGT, for precise prescreening of germline BRCA1/2 mutations in high-risk breast cancers.
- To integrate histopathological features from whole-slide images with clinical phenotypes from electronic health records for improved mutation prediction.
- To establish a cost-effective and scalable AI-driven approach for hereditary breast cancer prescreening.
Main Methods:
- MAIGGT utilizes a multimodal deep learning architecture, incorporating a Transformer-based generative model.
- The framework integrates features from whole-slide histopathology images and electronic health records.
- A cross-modal latent representation unification mechanism captures complementary insights from the integrated data.
Main Results:
- MAIGGT demonstrated robust performance across three independent cohorts, achieving areas under the ROC curve of 0.925, 0.845, and 0.833.
- The model outperformed single-modality approaches in predicting germline BRCA1/2 mutations.
- Interpretability analysis revealed distinct tumor microenvironment patterns associated with BRCA1/2 mutations, including inflammatory infiltration and nuclear heterogeneity.
Conclusions:
- MAIGGT offers a novel, accurate, and interpretable method for prescreening germline BRCA1/2 mutations in high-risk breast cancer.
- The integration of digital pathology and clinical data via AI represents a significant advancement in hereditary cancer risk assessment.
- This AI framework has the potential to improve the accessibility and efficiency of genetic testing in routine clinical practice.
