Adaptive Fusion of Radiomics and Deep Features for Lung Adenocarcinoma Subtype Recognition

• DOI: 10.48550/arxiv.2308.13997
• Publication Date: 2023-01-01
• Authors: Jing Zhou; Xiaotong Fu; Xirong Li; Wei Feng; Zhang Zhang; Ying Ji
• Keywords: Discriminative model; Feature (linguistics)

The most common type of lung cancer, adenocarcinoma (LUAD), has been increasingly detected since the advent low-dose computed tomography screening technology. In clinical practice, pre-invasive LUAD (Pre-IAs) should only require regular follow-up care, while invasive (IAs) receive immediate treatment with appropriate cancer resection, based on subtype. However, prior research diagnosing mainly focused classifying Pre-IAs/IAs, as techniques for distinguishing different subtypes IAs have lacking. this study, we proposed a multi-head attentional feature fusion (MHA-FF) model not from Pre-IAs, but also IAs. To predict subtype each nodule accurately, leveraged both radiomics and deep features extracted images. Furthermore, those were aggregated through an adaptive module that can learn attention-based discriminative features. utility our method is demonstrated here by means real-world data collected multi-center cohort.