FDG PET-CT combined with deep learning has demonstrated significantly improved survival prediction in non-metastatic breast cancer, highlighting the potential of advanced imaging analytics to refine prognostic assessment and clinical decision making.
Expanding the Role of FDG PET-CT in Breast Cancer
Fluorine 18 fluorodeoxyglucose positron emission tomography computed tomography (FDG PET-CT) is increasingly used across the breast cancer pathway, including initial staging, assessment of recurrence, evaluation of treatment response, and follow up. In non-metastatic disease, where patients may have nodal involvement but no distant metastases, quantitative measures derived from the primary tumour have already shown prognostic relevance. However, most existing approaches rely on predefined metrics such as maximum standardised uptake value, metabolic tumour volume, and total lesion glycolysis. While clinically useful, these parameters may fail to capture the complex and high dimensional information embedded within FDG PET-CT imaging data.
Deep Learning for Prognostic Stratification
To address these limitations, investigators conducted a deep learning-based analysis of FDG PET-CT in a large retrospective cohort of patients with non-metastatic breast cancer. Prognostic value was assessed from multiple perspectives, moving beyond visual interpretation alone. The researchers developed a multi omics prognostic stratification model that integrates clinical data, FDG PET-CT images, and corresponding medical reports. Using cross modal attention and transformer-based architectures, the model was designed to predict overall survival and disease-free survival with greater accuracy than single modality approaches.
Interpretability and Clinical Performance
To enhance clinical applicability, interpretability was embedded into the model. This included causal explanations, visualisation-based insights, and semantic interpretations aimed at supporting clinician understanding and transparent use of predictions. The multi omics prognostic stratification model demonstrated markedly improved performance compared with single omics models, tumour node staging, and molecular subtyping. Predictive accuracy was strong for overall survival and disease-free survival, with C index values for overall survival of 0.75 (95% CI: 0.69–0.81) and for disease free survival of 0.71 (95% CI: 0.65–0.77). These findings indicate that integrating FDG PET-CT with deep learning and clinical data can substantially improve risk stratification, supporting more personalised prognostic assessment for patients with non-metastatic breast cancer.
Reference
Liang X et al. Multi-omics deep learning improves FDG PET-CT-based long-term prognostication of breast cancer. Npj Precision Oncology. 2026; https://doi.org/10.1038/.
