Automatic classification of fluorescence and optical diffusion spectroscopy data in Neuro-Oncology

摘要

The relevance of automatic classification in glioma neurosurgery is due to the high geterogenity of this type of tumors. Characteristic signs of tumor tissue (high proliferative status, hypoxia,high density of nuclei and cells, myelin destabilization) can be determined using optical spectroscopy techniques. But the tissue in the area of interest may possess not all distinctive features and a full-fledged intraoperative analysis must take into account their complex. Methods of optical spectroscopy allow detecting various diagnostically significant parameters non-invasively. 5-ALA induced protoporphyrin IX is frequently used as fluorescent tumor marker in neurooncology. At the same time analysis of the concentration and the oxygenation level of haemoglobin and significant changes of light scattering in tumor tissues have a high diagnostic value. This paper presents an original method for the simultaneous registration of backward diffuse reflectance and fluorescence spectra, which allows defining all the parameters listed above simultaneously. We have created a software and hardware, which allowed (as compared with the methods currently used in neurosurgical practice) to increase the sensitivity of intraoperative demarcation of intracranial tumors from 78% to 96%, specificity of 60% to 82%. In this article, on the basis of neurosurgical removal of glioblastoma multiform with spectroscopic guidance, conducted in the N.N. Burdenko National Medical Research Center of Neurosurgery, analysis of different algorithm of automatic classification of spectroscopic data was carried out. The result of analysis of different techniques of automatic classification shows that in our case the most appropriate is the k Nearest Neighbors algorithm with cubic metrics.