RESUMEN
This paper proposes a deep learning-based approach to detect COVID-19 infections in lung tissues from chest Computed Tomography (CT) images. A two-stage classification model is designed to identify the infection from CT scans of COVID-19 and Community Acquired Pneumonia (CAP) patients. The proposed neural model named, Residual C-NiN uses a modified convolutional neural network (CNN) with residual connections and a Network-in-Network (NiN) architecture for COVID-19 and CAP detection. The model is trained with the Signal Processing Grand Challenge (SPGC) 2021 COVID dataset. The proposed neural model achieves a slice-level classification accuracy of 93.54% on chest CT images and patient-level classification accuracy of 86.59% with class-wise sensitivity of 92.72%, 55.55%, and 95.83% for COVID-19, CAP, and Normal classes, respectively. Experimental results show the benefit of adding NiN and residual connections in the proposed neural architecture. Experiments conducted on the dataset show significant improvement over the existing state-of-the-art methods reported in the literature. © 2022 ACM.
RESUMEN
Using deep learning approaches, this work presents a fully automated system for diagnosing COVID-19 from volumetric chest computed tomography (CT) scans. Transfer learning technique has been used to detect and classify CT scan data into three categories: COVID-19, CAP (Community-acquired pneumonia), and normal cases. The proposed model was built on top of the pre-trained AlexNet model's architecture and was capable of performing multi-classification tasks with a promising accuracy of 98.03%. The results demonstrate that the proposed model outperforms other current models and may thus be utilized as a potential tool for COVID-19 patient diagnosis. © 2021 IEEE.