New AI framework enables reliable, cost-effective prediction of PIK3CA mutations in breast cancer

accurate detection PIK3CA Mutations are essential for personalizing breast cancer treatment, especially PI3K-targeted therapy. However, traditional molecular tests are not always available, and single-modality predictive models have limited performance. This study introduces a multimodal artificial intelligence framework that combines digital pathology and structured clinical data to provide a robust solution for prediction. PIK3CA Mutation status in breast cancer. The proposed deep learning model provides a cost-effective, scalable, and reliable alternative to traditional sequencing by integrating whole-slide imaging (WSI) and clinical variables, with excellent performance across multiple clinical datasets.

Breast cancer is one of the most common malignancies worldwide, and mutations in the PI3K/AKT/mTOR (PAM) signaling pathway are prevalent during its development. Of these, PIK3CA Mutations play a pivotal role in guiding treatments with PI3K inhibitors, which have shown promising antitumor effects. However, traditional molecular assays such as polymerase chain reaction (PCR) and next generation sequencing (NGS) require expensive infrastructure and are not always feasible in routine clinical practice. Deep learning models have emerged as a cost-effective solution to predict important mutations from digital pathology images. Nevertheless, most existing models rely on single-modal data and often lack the complementary insights that structured clinical data can provide. These challenges highlight the need for improved predictive models.

Research published in (DOI: 10.20892/j.issn.2095-3941.2025.0771) Cancer biology and medicine In February 2026, a team of researchers from the 4th Hospital of Hebei Medical University developed a new multimodal artificial intelligence (AI) model for prediction. PIK3CA Mutations in breast cancer. The model integrates deep learning-based analysis of whole-slide pathology images with structured clinical data such as age, molecular subtype, and lymph node status. This study utilized data from The Cancer Genome Atlas (TCGA) and three external clinical cohorts to demonstrate the robustness of this model and its potential as an accessible alternative to molecular testing in diverse clinical settings.

A multimodal framework for research, known as multimodal PIK3CA The model (MPM) combines two components: a histopathological model and a clinical model. The histopathology model uses a transformer-based pre-trained encoder (H-optimus-0) and a clustered constrained attention multiple instance learning classifier (CLAM-SB) to process high-resolution whole slide images. This model identifies morphological features associated with: PIK3CA mutation. Clinical models based on XGBoost analyze structured clinical data to predict mutational status. Both models generate independent probabilistic predictions that are fused using a decision-level late fusion strategy to generate the final mutation status prediction. MPM outperformed single-modality models, achieving an area under the curve (AUC) of 0.745 in internal testing and stable performance (0.695 to 0.680 AUC) across external validation datasets. Inclusion of clinical variables such as molecular subtypes and lymph node status improved the predictive accuracy of the model, highlighting the importance of combining morphological and clinical data. This study also demonstrated the ability of this model to generalize across a diverse cohort, making it a promising tool for real-world clinical applications.

“This multimodal AI framework represents a significant advance in computational pathology. By integrating complementary clinical and morphological data, our model not only enhances the prediction of pathological conditions,” said Dr. Yueping Liu, lead author of the study. PIK3CA It not only detects mutations but also provides a scalable and cost-effective solution for clinical practice. Strong generalizability across diverse cohorts has the potential to improve personalized treatment decisions for breast cancer patients and bridge the gap between advanced molecular tests and routine clinical workflows. ”

MPM’s robust performance and ability to incorporate both digital pathology and clinical data make it a valuable tool for supporting clinical decision-making. This model provides a practical and cost-effective alternative to traditional molecular testing, which is often inaccessible in resource-limited settings. MPM has strong versatility across different medical centers and patient cohorts and can be implemented in daily clinical practice to make predictions. PIK3CA Detect mutations in breast cancer and guide the use of PI3K-targeted therapies. Future research may focus on improving the model for other mutations and cancers and expanding its applicability to precision oncology.

###

References

Toi

10.20892/j.issn.2095-3941.2025.0771

Original source URL

https://doi.org/10.20892/j.issn.2095-3941.2025.0771

Funding information

This study was financially supported by Hebei Natural Science Foundation (Grant No. H2024206504), Hebei Provincial Medical Science Research Project (Grant No. 20260484, 20260530), and Central Universities Basic Research Fund (Grant No. 20822041J4123).

About Cancer biology and medicine

Cancer biology and medicine (CBM) is a peer-reviewed open access journal sponsored by the Chinese Anti-Cancer Association (CACA) and Tianjin Medical University Cancer Institute & Hospital. The journal provides innovative and important information monthly on the biological basis of cancer, the cancer microenvironment, translational cancer research, and all aspects of clinical cancer research. The journal also publishes important insights into cancer types indigenous to China. The journal is indexed in SCOPUS, MEDLINE, and SCI (IF 8.4, 5-year edition IF 6.7), and the full text is freely available to clinicians and researchers worldwide (http://www.ncbi.nlm.nih.gov/pmc/journals/2000/).