Location of Regions of Interest in Tepscan images: Using Entropy Thresholding Associated with a Direction Vector and Related Component Analysis

Positron emission tomography (PET) is a commonly used examination nowadays, especially in cancerology. Thus, many methods of segmentation of Regions Of Interest (ROI) on PET images have been proposed in the literature. Among these methods, we can note iterative approaches, considering the characteristics of the patient, others based on pattern recognition, watersheds, etc. These methods have one major inconvenience: they require a calibration step on each device and each PET image reconstruction method. One can also mention the great algorithmic complexity that they induce. The aim of this work is to highlight hypermetabolic foci, our ROIs. To this end, we present an adaptation of image segmentation by two-dimensional entropy maximization, based on "recuit microcanonique". The search for segmentation thresholds, to which we add a direction, is carried out in steps of decreasing energy. In this process, the computation time as well as the localization of the ROI improves. The algorithm is tested on Tepscan images in DICOM format and compared to images where the area of interest has been manually marked.