Bi-omics prognostic model for invasive ductal carcinoma using deep learning
Mingwei Zhang, Tian Qiu, Wenbing Yang, Zhipeng Zhang, Bokai Shi · 2023
Invasive ductal carcinoma is a common subtype of breast cancer, and current prognostic models are mainly single-omics models. We aimed to develop a bi-omics model combining cancer-associated fibroblast(CAF) gene expression and Whole-field digital slice image(WSI) features to improve the accuracy of predicting prognostic effect. A total of 491 patients were included in the experiment and were randomized into training and validation sets. CAF genes associated with patient survival were screened by gene set enrichment analysis (GSEA), differential expression analysis and Cox regression. Whole-field digital slice image features were extracted using deep learning-based U²Net model and artificial feature approaches, and a bi-omics prognostic model was established by multi-factor Cox regression, and finally survival analysis and AUC values were applied to validate the model accuracy. The prognostic prediction performance of the combined bi-omics model was significantly improved compared to the single-omics model. This study highlights the potential of integrating CAF gene expression and WSI features to better predict the prognosis of IDC patients and provide valuable information for clinical decision making and personalized treatment strategies.