RFSC: Multimodal medical image alignment fusion diagnostic classification network based on de discriminator image translation

Abstract

Lung cancer, one of the most prevalent and deadly cancers worldwide, poses a significant threat to human health. Computer-aided diagnosis systems currently play a critical role in lung cancer treatment. However, the diagnostic effectiveness of high-dose Computed Tomography (CT) methods is limited, also raising concerns about radiation exposure. Furthermore, standalone Magnetic Resonance Imaging (MRI) often fails to depict lesion contours accurately, thereby restricting its usability in computer vision for lung nodule diagnosis. In light of these challenges, studies have proposed a method that combines low-dose CT and MRI images for lung nodule diagnosis. Nevertheless, the efficacy of this fusion method has attracted attention among researchers. To enhance the efficiency of image fusion, this study introduces a jointly trained multimodal registration network that maximizes the preservation of image information. Concurrently, the registered images are utilized for image fusion, and a classifier based on lung nodule image features is constructed. Finally, multiple models are integrated into a multitask artificial intelligence diagnosis model to enhance clinical diagnostic efficacy. Compared to other registration models, the registration Dice Similarity Coefficient (DSC) achieves 0.9165591, and the accuracy of benign-malignant lung nodule classification reaches 89.916%.