An Optimal and Automatic Graph Cut Method for Biomedical Images Using Compactness Measure

This work aims to achieve an automatic and optimal graph cut phase based on a segmentation method presented in a previous paper. A graph-based segmentation algorithm, starting from a seed point belonging to the region of interest (ROI), is able to find the Minimum Path Spanning Tree (MPST) by using a new cost function and an optimal aggregation criterion. In order to extract the ROI, a graph-cut of the obtained tree is absolutely necessary. By definition, the main drawback of the graph-based segmentation methods is the loss of spatial and contextual information. To overcome this problem, a new method based on compactness measure is here proposed The present approach is applied to the biomedical field, considering Magnetic Resonance Imaging (MRI) volumes of the hand and neurological districts.