A contrastive learning-based neural network to synthesize cell subpopulation features from DCE-MRI for predicting prognosis in breast cancer

Breast cancer is characterized by genetic heterogeneity, exhibiting diverse gene expression profiles within individual tumors. Despite the critical insights offered by genomic analyses in delineating this complexity, the invasive and costly nature of such examinations limits their widespread application in clinical settings. Magnetic resonance imaging (MRI) is noninvasive and widely used for cancer diagnosis and treatment. In view of this, we developed a contrast learning-based framework to synthesize genomic characteristics from MRI. Specifically, we extracted image features using the 3D-ResNet18 architecture, while cell subpopulation features were obtained through a multilayer perceptron (MLP). The contrastive learning network aligns the image features and genomic features in the representation space using a contrastive loss. We saved the weights of the image feature extractor from the contrastive learning stage as pretraining weights for the generator in the generative model and used the discriminator to distinguish between the generated and the real immune cell subpopulations. Further survival analysis of the generated immune cell subpopulations was conducted using the log-rank test. The dataset consisted of 135 patients, with 81 samples allocated to the training set and 54 samples assigned to the testing set. Based on the univariate Cox hazard model, 10 immune cell subpopulations significantly associated with overall survival were identified. Immune cell subpopulations were generated using the proposed contrastive learning-based generative model, and the risk score was calculated using multivariate Cox regression. The generated risk scores of immune cells achieved the R square of 0.48 and 0.43 in the validation and test sets, respectively. Significant differences in prognosis were observed after grouping the patients according to risk scores, with p values of 0.033 and 0.011 in the validation and test sets, respectively.