Using Bayesian Network to Integrate CT Image Data and Epigenomic Data for Tumor Classification

With the development of technology, image data (such as CT and MRI) and epigenomic data (such as RNA-seq, miRNA, and Methylation) are increasingly used for disease detection and prevention, especially for cancer. Due to the high dimension of the dataset and different feature space among different types of the datasets, prediction using the combination of multiple types of data is very challenging. We proposed a new method that utilized SFPCA, 3-D FPCA, sparse SDR, and SVM to build a prediction model within each dataset, and used Bayesian Network(BN) to integrate multiple types of datasets for classification and prediction. We applied the method to predict tumor existence for ovarian cancer. The results showed that the average prediction accuracy for each individual dataset is 74% for the CT image dataset, 69% for the mi-RNA dataset, 72% for the DNA Methylation dataset. After using Bayesian Network to combine different types of the datasets, the average prediction accuracy increased to 79%