A Novel Radiogenomics Framework for Genomic and Image Feature Correlation using Deep Learning

Precision medicine still remains to be a prevalent treatment strategy which has been continuously pushed forward by the upcoming targeted therapies. To improve the precision and quantitative level, researches in radiomics and radiogenomics have devoted much of their endeavors to transform digital standard of medical images to mineable high-dimensional data by way of extracting mathematically quantitative features. However, most of the prior efforts could not effectively combine multi-source medical data sets together to generate satisfactory results and then visualize diagnoses by unifying low level features from images and other sources. In this paper, we design a novel and meaningful framework in order to map the features between medical images and gene expression profiles and quantity their correlations. To ameliorate, we take full advantage of deep learning methods, and characterize the lung cancer clinically at both genome and image levels. Our newly-devised protocol could give a strong association between gene and tumor growth statues, furthermore, it could provide cogent visual results for clinical research directly. The research presented in this paper could provide more comprehensive characterizations of tumor phenotypes, statues, and outcomes. As a result, it may be noted that, all of our prior efforts could contribute to the bigdata analysis for biomarker signatures, images, and 'Omics'.