A Dense Layer Stacked CMT-DenseNet Fusion Model for COVID Detection

Advancements in deep learning and transformers have resulted in the development of Vision Transformers and Convolutional Neural Networks Meet Vision Transformers (CMTs). Though CMTs capture long-range dependencies as well as local knowledge inside the image patches, there is a need for a mediator which serves as a strong learner to combine local and global knowledge components of inputs more efficiently. In this paper, we present a fusion model of DenseNet with dense layer stacked CMT to improve CMT’s performance by introducing dense layers within CMT. These dense layers serve as a mediator that combines the two knowledge components more efficiently. The experimental results on SARS COVID 2 CT SCAN dataset demonstrate the efficacy of the proposed fusion model. Our model shows an improved accuracy of 96.99% when compared to the individual models (that is, DenseNet or CMT), which indicates that the collective knowledge of DenseNet complements the CMT model well for medical images.