Can Radiomic Features Boost The Performance Of Deep Learning Upon Histology Images?

Radiomics,as a promising technique for feature extraction,can capture a wealth of information from multi-modal medical images through highthroughput feature representations.This confers the advantage of interpretability that deep learning(DL)features do not have,and that’s the reason why it’s widely being used in computer-aided systems to provide additional expert support.In this paper,we try to explore whether the different nature of radiomics features,compared to DL features,can provide complementary information for a DL diagnosis system and thus boost its performance.A DL framework successfully used for cancer histology image classification works as the baseline to mirror the influence brought by importing radiomics features as complementary information to the DL system.The result shows that the system owning complementary information given by radiomics features performs better compared with the system purely using DL features.