Developing Poly-Mask for Automatic ROI Selection of Medical Images

Abstract

Region-of-Interest (ROI) or Region-of-Non-Interest (RONI) selection is a key step in medical image (MI) watermarking to avoid any concern of possible incorrect diagnosis from watermarked MIs. Generally, an automatic ROIs/RONIs selection is an ever challenging task due to the variety of locations and shapes of ROIs/RONIs in different modality MIs. In this paper, we have developed and presented Poly-Mask, an algorithm for automatic selection of ROIs. Poly-Mask takes an input MI, processes it morphologically and computes a polygonal mask, poly-ROI that captures the ROIs by using convex-hull without consideration of any basic shape used in earlier methods. A minimum information (e.g., values of vertices, minimum and maximum axes) is then determined for directly regenerating the poly-ROI later for watermark embedding and extraction. Experimental results demonstrate that the proposed Poly-Mask algorithm can select and accurately generate the poly-ROIs and so the ROIs/RONIs of different MIs, which can be used for different MI watermarking applications.