Total Variation Distance-Enhanced Selective Segmentation for Medical Images

Abstract

In this paper, we propose an enhanced local intensity clustering energy functional designed for selective segmentation of medical images, particularly those affected by intensity inhomogeneity. The functional includes an area constraint term based on a total variation (TV) distance function derived from the single-scale Retinex output image. This TV distance function measures an unusual distance between points in the image domain and specified marker points, ensuring accurate localization of the selected objects. By combining this with local intensity clustering fitting energy and contour length regularization, the resulting minimization model achieves precisely selective segmentation and tight object wrapping. Moreover, instead of solving the Euler-Lagrange equation or using the level set method, we introduce an efficient iterative convolution-thresholding method to implement the model numerically. This method guarantees energy decay and enables faster convergence to a stable partition. Numerical experiments on some medical images demonstrate the effectiveness and efficiency of our proposed approach for selective image segmentation.