VLSI Implementation of Optimized CNN based COVID-19 Lung Infection Segmentation and Classification from CT Image

The outbreak of COVID-19 challenged the existence of human life on earth. The diagnosis and treatment of this disease is highly crucial in current scenario. Since there is low difference in the intensity of normal cells and affected cells, Computed Tomography (CT) is an efficient tool for the diagnosis of lung infections caused due to COVID-19. In order to train a deep neural network, there is a requirement for huge number of labelled images. There is a necessity to develop an efficient neural network that requires less number of training images. To overcome these problems a novel network is developed for lung CT segmentation (SqueezeNet). For the extraction of energy values from the segmented image, Discrete Wavelet Transform (DWT) with lifting scheme is incorporated in the framework. These energy values are used for training the classifier (ResNet). The entire framework is implemented on hardware using Virtex 2 Pro FPGA. The performance of the proposed system is evaluated using Mean Absolute Error (MAE) and Specificity. The MAE of the system is found to be 0.099, which is very low compared to existing classifiers. The specificity of the system is 0.978, which is higher than that of existing classifiers.