RCANet: A Deep Learning Model for Thoracic Diseases Classification of Chest X-ray Images

In recent years, artificial intelligence, especially machine learning, has received much attention in the field of medical imaging. Deep convolutional neural networks rely on large amounts of data during training, but the actual chest X-ray images dataset has substantially fewer positive examples than negative ones. When training the model with unbalanced data, the algorithm prioritizes the majority class, making the disease classification model heavily biased. The commonly used networks such as Resnet-50 and DenseNet-121 can improve the redundant use of information in the process of feature extraction and the problem of insufficient discriminative power in some medical images discriminative tasks by introducing the attention mechanism module. However, the widely used SE attention mechanism focuses only on the attention of the channel and has limited ability to increase the model's ability to extract truly useful key features. We propose a weakly supervised learning model combining a coordinate attention mechanism, and experiments on the Chest X-ray14 dataset yield significant improvements over previous networks.