A New Classification Method for COVID-19 CT Images Based on Transfer Learning and Attention Mechanism

Abstract

Nowadays, the typical tools employed in the diagnosis of the pandemic coronavirus disease, COVID-19, including Real-Time Reverse Transcription Polymerase Chain Reaction (RT-PCR) and Polymerase Chain Reaction (PCR), which are less sensitive, time-consuming, and demanding assistance from expert medical personnel assistance. Computed tomography (CT) as the artificial intelligence (AI) technological utilized in high accurate COVID-19 infection screening in a short amount of time is tremendously helpful. To address those limitations mentioned above, In this paper, a robust, optimized model for detection of the COVID-19 automatically in digital CT images is proposed utilizing the technique based on transfer learning and attention mechanism derived from deep learning. MobileNet-V1 architecture of transfer learning was applied to make the model more lightweight and reduce the computation, while setting to be the pre-trained mode meanwhile. In addition, the attention mechanism of SENet’s called Squeeze-and-Excitation (SE) module was employed to let it learn the significance of various channel features automatically. Two experiments, with transfer learning and attention mechanism technique or not, were employed to assess the function of the model. Noteworthy, the accuracy, precision, recall, and F1-score were 95.97%, 93.47%, 94.27%, and 94.01% respectively. The results reveal that the optimized approach outperform the comparative models.