An artificial intelligence model leveraging radio-multiomic analysis of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) shows promise in predicting breast cancer outcomes and treatment response. The study, published in Nature, demonstrates that the AI can accurately forecast relapse-free survival (RFS) and overall survival (OS) by integrating radiomic features extracted from MRI scans with multiomic data. This approach offers a more comprehensive understanding of tumor biology and could lead to more personalized treatment strategies.
AI Model Development and Validation
The AI model was developed using data from the Fudan University Shanghai Cancer Center (FUSCC) cohort and validated on two independent cohorts: the DUKE University dataset and the I-SPY1 dataset. The FUSCC cohort included 466 patients with prognosis information and 420 with multiomic data. The DUKE cohort comprised 619 patients for prognosis prediction and 217 for treatment response, while the I-SPY1 cohort included 128 patients for prognosis and 125 for treatment response.
The model identifies a 13-feature radiomic signature associated with prognosis by using Lasso and Cox regression. The prognostic power of the risk score from the radiomic signature was tested in the testing set of the FUSCC cohort, the DUKE cohort, and the I-SPY1 cohort. Multiomic data analysis was performed in the FUSCC cohort to explore the biological foundation of the prognostic radiomic signature. The radiomic risk score was transferred to the DUKE cohort and I-SPY1 cohort as a radiomic model to predict neoadjuvant chemotherapy response. Clinical and combined clinic-radiomic models were also built to predict treatment response.
Radiomic Feature Extraction and Analysis
Researchers extracted radiomic features from tumor, peritumoral, and whole tumor regions of interest (ROIs) on DCE-MRI scans. These features included spatial domain characteristics (first-order, textural, and wavelet features) and sequential features, which capture dynamic changes in the tumor over time. The PyRadiomics package was used for feature extraction, and rigorous quality control measures, including inter- and intra-observer reproducibility tests, were implemented to ensure data reliability.
Integration with Multiomic Data
To understand the biological basis of the radiomic signature, the researchers integrated it with multiomic data, including gene mutation, copy number variation, transcriptomics, proteomics, metabolomics, and pathomics data from the Chinese Breast Cancer Genome Atlas (CBCGA) project. Gene set enrichment analysis (GSEA) was performed to identify pathways associated with high- and low-risk groups, and differential abundance (DA) scores were calculated to assess changes in metabolite levels.
Predicting Treatment Response
The study also explored the ability of the radiomic signature to predict response to neoadjuvant chemotherapy. In the DUKE and I-SPY1 cohorts, the radiomic risk score, along with clinical factors, was used as input for a logistic regression model. The model's performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC).
Clinical Implications
"Our findings suggest that AI-powered radio-multiomic analysis can provide valuable insights into breast cancer prognosis and treatment response," said Dr. [Author name], lead author of the study. "By integrating imaging data with genomic and metabolic information, we can develop more accurate predictive models and potentially personalize treatment strategies for patients."
The AI model offers a non-invasive method to refine risk stratification and treatment planning in breast cancer, potentially improving patient outcomes. Further research is needed to validate these findings in larger, prospective clinical trials and to explore the model's utility in diverse patient populations.
