Automated Diagnosis of Pulmonary Nodules in 3D PET/CT Images Using Dual-path Densely Connected Networks with Cross-modal Fusion
This repository contains the code and data for the paper "Automated Diagnosis of Pulmonary Nodules in 3D PET/CT Images Using Dual-path Densely Connected Networks with Cross-modal Fusion". The code will be made public once the article is accepted！
PET/CT provides robust evidence for the automated diagnosis of benign versus malignant pulmonary nodules. However, inherent differences between imaging modalities have limited exploration of cross-modal interactions, hindering the effective integration of their complementary features. To bridge this gap, we propose a novel classification network based on wavelet convolution feature fusion, aiming to fully leverage the complementary properties of multimodal medical imaging information for pulmonary nodule diagnosis. Specifically, a customized Dense-HGMC module extracts multi-scale contextual features, while the DDGR module captures fine-grained lesion morphological characteristics. Additionally, we designed a wavelet-based feature fusion module to efficiently integrate multimodal features. Performance evaluation was conducted on a clinically recruited dataset comprising 812 paired nodule images. Results demonstrate that the model achieves state-of-the-art performance, surpassing existing fusion approaches and fully validating the effectiveness of this approach for precise lung nodule diagnosis.
The recruited dataset:
The experiments were conducted using Python version 3.8 (Python Software Foundation, Wilmington, DE, USA), PyTorch 2.0.0, with Compute Unified Device Architecture (CUDA) version 11.8, and an NVIDIA GeForce RTX 4090 GPU, running on Ubuntu 20.04 (Canonical, London, UK).