Boosting Advanced Nasopharyngeal Carcinoma Stage Prediction Using a Two-Stage Classification Framework Based on Deep Learning

Abstract Nasopharyngeal carcinoma (NPC) is a popular malignant tumor of the head and neck which endemic in world, more than 75% NPC patients suffer from locoregionally advanced nasopharyngeal (LA-NPC). The survival quality these depends on reliable prediction stages III IVa. In this paper, we propose two-stage framework to produce classification probabilities for predicting preprocessing MR images enhance further analysis. stage one transfer learning used improve effectiveness efficiency CNN models training with limited images. Then two output are aggregates using soft voting boost final prediction. experimental results show quite effective, performance perform better basic model, our ensemble model outperforms single as well traditional methods, including TNM staging system Radiomics method. Finally, accuracy boosted by is, respectively, 0.81 , indicating that method achieves SOTA LA-NPC addition, heatmaps generated Class Activation Map technique illustrate interpretability models, their capability assisting clinicians medical diagnosis follow-up treatment producing discriminative regions related Graphic