COVID Detection from Multimodal Images Using ELM and UNET

Abstract

Coronaviruses are serious illnesses that affect both people and animals. Currently, the novel COVID-19 coronavirus is rapidly spreading over the world, putting billions of people's health. The majority of COVID-19 patients had a lung infection, according to clinical examinations. Computed Tomography (CT) is an effective imaging tool for identifying lung disorders which can have more detailed information about chest region. Chest X-ray is more widely available due to its faster imaging time and lower cost. Deep learning, one of the most effective AI technologies, helps radiologists to analyze vast numbers of chest images, which is critical for rapid and reliable COVID-19 screening. The goal of this project is to create a new deep anomaly detection model that can be used quickly, reliable screening of COVID-19 from CT and X-Ray Images. Here, to create the segmentation map, UNET is used after training with standard database of CT and X-ray images of COVID effected peoples. To classify normal, COVID or pneumonia, Extreme Learning Machine (ELM) is used with features such as center symmetric local binary pattern (CSLBP), shape features and arithmetic features. To improve the performance, image denoising, smoothing, and normalization techniques are used in the pre-processing stage. To evaluate the performance, sensitivity, specificity, accuracy, and precision statistical measures were utilized. This work achieved a maximum accuracy of 99.96%. The results suggest that this work performs better than conventional covid classification works.