1 October 2022 in Bulletin of Electrical Engineering and Informatics
In response to the growing threat posed by COVID-19, several initiatives have been launched to develop methods of halting the progression of the disease. In order to diagnose the COVID-19 infection, testing kits were utilized; however, the use of these kits is time-consuming and suffers from a lack of quality control measures. Computed tomography is an essential part of the diagnostic process in the treatment of COVID-19 (CT). The process of disease detection and diagnosis could be sped up with the help of automation, which would cut down on the number of exams that need to be carried out. A number of recently developed deep learning tools make it possible to automate the Covid-19 scanning process in CT scans and provide additional assistance. This paper investigates how to quickly identify COVID-19 using computational tomography (CT) scans, and it does so by using a deep learning technique that is derived from improving ResNet architecture. In order to test the proposed model, COVID-19 CT scans that include a patient-based split are utilized. The accuracy of the model’s core components is 98.1%, with specificity at 97% and sensitivity at 98.6%.
11 May 2020 in IEEE Access
A Non-Uniform Sampling (NUS) methodology is presented, which improves the Angle of Arrival (AoA) estimation accuracy. The proposed sampling methodology is applied to extract the collected data efficiently and to reduce the possible dependency within rows/columns of the Covariance / Correlation Matrix (CM). The new sampled matrix approach is utilized to form such a projection matrix in order to provide better resolution of angles of arrival for closely incident signals on the antenna array receiver and to increase Degrees of Freedom (DOFs) compared to the classical criterion. The new direction-finding method based on the NUS methodology is called a Non-Uniform Projection Matrix (NUPM). A theoretical analysis is presented to demonstrate the advantage of the NUS methodology in terms of the obtained energy of the eigenvalues that are associated with the signal eigenvectors. It is proved that the proposed matrix sampling approach can provide better estimation resolution and detection of higher numbers of angles of arrival compared to the classical sampling method, without increasing array aperture size and the complexity of computations. A reduced-dimension process is applied to the NUPM in order to decrease the computational burden at the grid-scanning step. A computer simulation, including many scenarios, is implemented to justify the expected improvements of the NUS methodology compared to the classical approach. It is found that the new sampling distribution enhances noise immunity and increases DOFs compared to the conventional criterion. The performance of the NUPM method is compared with several AoA methods, including the Cramer-Rao Lower Bound (CRLB), and the obtained results verify the effectiveness and robustness of the proposed NUPM technique.