Non-Small Cell Lung Cancer Staging Using Deep Restricted Boltzmann Machine

Accurate stage prediction of Non-small cell lung cancer (NSCLC) is very essential for making the right treatment decisions. Radiomics is currently one promising means to capture intra-tumor heterogeneity and therefore to predict the cancer stage. In order to improve the prediction accuracy of radiomics for NSCLC stages, we proposed to use the deep restricted Boltzmann machine (DRBM) to replace the traditional classifiers used in the framework of radiomics. The experiments were performed with the datasets provided by MAASTRO Clinic in Maastricht. The experimental results show that the area under curve (AUC) of DRBM was increased by about 19% compared traditional logic regression and support vector machine methods, especially for stage I and stage IIIa, our proposed method is much better than the others.