ABSTRACT
Segmentation is the process of delineation of the desired region of interest (ROI) in medical images. The fuzzy c-means (FCM) is a classical data mining algorithm that gains importance in medical imaging for the extraction of ROI. The traditional FCM is subtle to noise and gets stuck at local minima. An improved FCM based on Gaussian kernel with crow optimization was proposed in this work for ROI extraction in corona virus disease (COVID-19) CT images. The CT imaging modality is the primary screening tool for COVID-19 disease. The median filter was used for preprocessing and the Euclidean distance in classical FCM clustering was replaced by the Gaussian kernel. The objective function was modified by the inclusion of crow search optimization (CSO) for cluster centroids selection. The optimization technique was used for random initialization of cluster centroids. The performance validation by metrics reveals the efficiency of the proposed segmentation model when equated with the classical FCM clustering and adaptive regularized kernel FCM clustering techniques. The incorporation of median filtering makes the clustering technique less sensitive to noise and the Gaussian kernel with CSO for cluster centroids initialization generates efficient segmentation results. The algorithms are developed in Matlab 2015a and tested on CT images.
