Multi-label classification of medical images based on feature fusion
Yiqi Zhu · 2025
This article considers the role of multi-label medical image classification and the related revolution in clinical diagnosis. Here, the use of deep learning technologies brings up the problem of single-label classification methods that may not be able to deal with the multi-label one as they do not cover the complex, sometimes coexisting, multiple-pathological features in medical images. To address these challenges, we propose the CTTransNet speculative model, a combination of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), which aims to better integrate the local features and the global context. Similarly, a Transformer decoder is used to model inter-label dependencies, and a less-weighted fusing loss mechanism is utilized to deal with the data imbalance issue. The AUC of 81.68% on the ChestXray14 dataset is achieved by the proposed method, which shows its higher accuracy and superiority to the robustness compared to the traditional methods in multi-label medical image classification. This breakthrough thus brings about the complete information needed for a clinical diagnosis and makes it easier for personalized precision medicine to develop.