Interpretative computer-aided lung cancer diagnosis: From radiology analysis to malignancy evaluation

Background and Objective:Computer-aided diagnosis (CAD) systems promote effectiveness alleviate pressure of radiologists. A CAD system for lung cancer includes nodule candidate detection malignancy evaluation. Recently, deep learning-based pulmonary has reached satisfactory performance ready clinical application. However, evaluation depends on heuristic inference from low-dose computed tomography volume to malignant probability, which lacks cognition. Methods:In this paper, we propose a joint radiology analysis network (R2MNet) evaluate the via characteristics analysis. Radiological features are extracted as channel descriptor highlight specific regions input that critical In addition, model explanations, channel-dependent activation mapping visualize shed light decision process neural network. Results:Experimental results LIDC-IDRI dataset demonstrate proposed method achieved area under curve 96.27% AUC 97.52% explanations CDAM proved shape density were two factors influence be inferred malignant, conforms with cognition experienced Conclusion:Incorporating evaluation, diagnostic procedure radiologists increases confidence results. Besides, interpretation DNNs focus when they estimate probabilities.