New machine learning based method for identifying objects in biomedical images obtained by photon counting detectors.

Abstract

The article discusses a new approach to the problem of recognition, or rather the identification of objects by the shape of their intensity in images formed by photon counting detectors or those close to them. The main problem analyzed within the framework of the approach proposed is related to the quantitative estimation of similarity between the form of radiation intensity of the object in the image and the forms of intensity of previously observed objects (precedents), providing that all of them are given by the registered sets of discrete photocounts (~ photons). It is shown that when the intensity shape is approximated by a mixture of components of an exponential family, for the implementation of the approach proposed in this work a recurrent PM algorithm for estimating the parameters of the intensity shape from a finite sample of counts associated with radiation can be synthesized. The PM algorithm turns out to be a close analogue of the well-known K-means algorithm, which is one of the most popular in the field of machine (statistical) learning. The main features of the method and its implementation are illustrated by the example of applying the method to the problem of image fragments identification for the COVID ‒ CT ‒ Dataset database.