A Novel Approach to Detect COVID using DenseNet Architecture
S. Senthil Pandi, K. Deepak Kumar, A. Senthilselvi, D. Roja Ramani
Abstract
Evolution of machine learning and deep learning approaches, and novel anatomical and functional imaging modalities have resulted in several computer-aided diagnosis and detection systems. These systems are centered around pathology localization, detection and classification of abnormalities. Conventional medical image analysis approaches such as classification and segmentation are tailored to the problem of detecting or classifying abnormalities in pathology images. Machine learning and deep learning models are trained on a large number of pathology images in the form of labeled image datasets, so that they generalize well with unseen data. A binary classifier model which is trained on CT images which are quality enhanced with optimal filters is proposed for COVID19 detection from chest images. This paper is based on DenseNet architecture featuring dense connections, to propagate and concatenate feature maps across dense layers for collective learning. This paper captures diverse manifestations of COVID19 specific infections from CT images, improving the detection accuracy. Further, explainable analysis of this paper provides insights on the morphologies of the infection patterns for extensive analysis.