Abstract: The pandemic coronavirus disease-2019 (COVID-19) has infected millions of people in over 200 countries and territories as of 2021. It is very necessary to detect COVID-19 in the initial stage to provide appropriate medical treatment to patients and also to protect the uninfected people. For this reason, we develop a framework to automatically segment COVID-19 CT images using k-means clustering and use them to train proposed convolutional neural network to classify COVID-19 from normal CT images. Rapid growth in machine learning and deep learning has been doing great work to reduce time of radiologists by assisting them in the diagnosis of COVID-19. Our framework is evaluated upon 349 positive and 397 negative CT scans to detect COVID-19 and help in taking appropriate diagnostic decisions. To evaluate the performance of proposed approach, we compared our results with pre-trained models such as VGG19, InceptionV3 and ResNet50.