Integrative Radiogenomics Using MRI Radiomics and Microarray Gene Expression Analysis to Predict Pathological...

Given the variable pathological complete response (pCR) rate of neoadjuvant chemotherapy (NAC) in patients with breast cancer, identifying predictive markers is crucial. This study evaluated the predictive accuracy of three machine learning-based models: (1) radiomics using MRI features; (2) genomics based on DNA microarray data; and (3) radiogenomics integrating both MRI and microarray data to predict pCR after NAC across all breast cancer subtypes. This study aimed to determine which model provides the most precise non-invasive prediction by utilizing a consistent dataset and analytical pipeline.

In this retrospective study, 112 patients with breast cancer who underwent DNA microarray analysis and dynamic contrast-enhanced MRI before receiving NAC at a single institution between July 2006 and November 2016 were classified into pCR (N = 21) and non-pCR (N = 91) groups. The prediction accuracy of pCR after NAC was evaluated for three models using repeated stratified nested cross-validation (CV). Model performance was assessed by the area under the receiver operating characteristic curve (ROC-AUC), and statistical significance was tested using DeLong's test.

Among the 112 patients, the radiogenomics model yielded an AUC of 0.607 (95% confidence interval (CI): 0.438-0.758), outperforming both the radiomics (AUC 0.563, 95% CI: 0.410-0.718) and the genomics (AUC 0.559, 95% CI: 0.379-0.722) models. However, this improvement was not statistically significant (p>0.05).

Machine learning-based radiogenomics, which combines MRI features and DNA microarray data, improved the accuracy of pCR prediction after NAC, although the improvement was not statistically significant. These findings suggest the potential utility of radiogenomics as a non-invasive tool to support treatment decision-making in patients undergoing NAC.