Design of COVID19 Detection for Risk Identification using Deep Learning Approach

• Published in: 2023 International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) pp. 1 - 5
• Main Authors: Chaturvedi, Abhay, J, Rohan, Saktheeswaran, Murali, J, Samson Isaac, Kathan, Ambujam, Sekhar, G.G. Raja
• Format: Conference Proceeding
• Language: English
• Published: IEEE 29.04.2023
• Subjects: Computed tomography; Computer viruses; COVID-19; COVID-19 deep learning algorithm; Deep learning; Image segmentation; Preprocessing; Segmentation; Support vector machines; Visualization
• DOI: 10.1109/ICDCECE57866.2023.10150731

Coronavirus infection is caused by one of the global issues of a deadly virus. The letters 'CO' stands for different types of symptoms, 'VI' for infection easily spread, and 'D' for everybody. Detection and diagnosis are based on various symptoms, and because the virus primarily affects the lung area of the human body, most people test their health condition using CT scans (Computed Tomography). The existing method Support Vector Machine Classifier (SVM)-based image processing technique, identifies CT (Computed Tomography) lung images from patients, which has drawbacks such as a lower positive rate and a less accurate affecting region. So, in the proposed method, the lung CT (Computed Tomography) image is used to classify deep learning. Preprocessing, Segmentation, and Classification are all steps in the digital image processing process. The first step image is obtained by taking the input CT (Computed Tomography) image and processing it into various regions while obtaining the deep level of the specified area using preprocessing bilateral filter. The second step is segmentation using K-means clustering the image segments into different regions or objects like effect and normal regions. This main object is removing the unwanted area. The third step Classification method is based on deep learning. The image is classified with different mathematical functions and identified as the extract part of the lung image. Output results are distinguished sickness and non-illness areas in the lung. The difference in the dark level of each locale is recorded and used as a highlight to identify infection areas identified by low fluctuation contrast with non-COVID-19 affected areas. Simulation is used to validate the proposed method's performance. The proposed method has an accuracy of 84.35% and a false rate of 15.65%.