Image segmentation from noisy CT image has a number of applications in different clinical diagnosis. In order to automatically perform this task from an under-sampled noisy CT image, we have merged compressed sensing reconstruction technique and hierarchical clustering algorithm together in this paper. Denoising based approximate message passing (D-AMP) algorithm is used as a compressed sensing reconstruction technique. Then the unlabeled local feature points from the reconstructed image are classified into two segments using hierarchical clustering algorithm. We have evaluated the performance of both these techniques by using a number of metrics. The research knowledge from this project might be extended in future works such as reconstruction of CT image from raw medical data and 3D segmentation of medical images.