Prognostic Survival Prediction Model of Non-small Cell Lung Cancer Based on Multimodal Feature Fusion

Non-small cell lung cancer accounts for a large percentage of all lung cancers. Given the diversity of therapeutic modalities and prognostic performance, it is of great clinical importance to predict prognostic survival in NSCLC patients. In prognostic prediction, more and more studies are building predictive models based on multiple factors. Although the deep learning features can highly express the lesion status, there are few studies that directly employ this feature for prognosis prediction. Hence, a 2-year survival prediction model for non-small cell lung cancer based on MLP is proposed by integrating the radiomics characteristics and deep learning features extracted from CT images along with clinical analysis features. Furthermore, deep learning features are extracted from central tumor patches in axial, coronal and sagittal planes of CT images to deliver more comprehensive and precise features, using two-dimensional modeling and three-dimensional. Eventually, the proposed model achieves AUCs of 0.886 +/- 0.031 and 0.825 +/- 0.023 on NSCLC-Radiomics and hospital data sets, respectively.