Distinguishing COVID-19 early may help in conceiving a fitting treatment plan and disease containment choices. In this investigation, we show how move gaining from profound learning models can be utilized to perform COVID-19 discovery utilizing pictures from three most ordinarily utilized clinical imaging modes X-Ray, Ultrasound, and CT filter. The point is to give over-focused on clinical experts a second pair of eyes through wise profound learning picture arrangement models. We recognize a suitable Convolution Neural Network (CNN) model through beginning similar investigation of a few mainstream CNN models. We then, at that point upgrade the chose VGG19 model for the picture modalities to show how the models can be utilized for the exceptionally scant and testing COVID-19 datasets. We feature the difficulties (including dataset size and quality) in using current freely accessible COVID-19 datasets for creating useful deep learning models and what it unfavorably means for the teach ability of complex models. We likewise propose an image prehandling stage to make a dependable picture dataset for creating and testing the profound learning models. The new methodology is meant to decrease undesirable commotion from the pictures so that profound learning model scan center around recognizing illnesses with explicit highlights from them. Our outcomes show that Ultrasound images provide better recognition exactness thought about than X-Ray and CT examines. The test results high light that with restricted information, the greater part of the more profound organizations battle to prepare well and gives less consistency over the three imaging modes we are utilizing. The chose VGG19 model, which is then widely tuned with proper boundaries, acts in impressive degrees of COVID-19 discovery against pneumonia or normal for every one of the three lung picture modes with the accuracy of up to 86% for X-Ray, 100% for Ultrasound and84% for CT checks