Non-invasive diagnosis of Parkinson's disease based on skin autofluorescence spectra
Nikita P. Bainaev-Mangilev3, Vladimir V. Salmin1,2, Victor B. Loschenov3,4, Aryuna. B. Ochirova3, Maxim N. Andreev5, Ekaterina Yu. Fedotova5, Alla B. Salmina5, Sergey N. Illarioshkin5; 1Moscow Institute of Physics and Technology; 2Bauman Moscow State Technical University; 3National Research Nuclear University MEPhI; 4Prokhorov General Physics Institute of the Russian Academy of Sciences; 5Research Center of Neurology
Abstract
Parkinson's disease (PD) is the second most common neurodegenerative disease, affecting approximately 1% of people over 60. Currently, there are no effective treatments for PD. Nevertheless, early diagnosis of the disease makes it possible to provide decent conditions for patients and suppress the intensity of symptoms. Wu D. et al. showed the possibility of using autofluorescence spectra of skin and nail plates as a predictor of disease. Presumably, the effect is associated with the accumulation of the protein alpha-synuclein in the skin, in particular in the epidermis, which leads to changes in its spectral characteristics.
In this work, a method for processing and classifying the obtained skin autofluorescence spectra was developed and clinical trials were conducted. The experimental group included 54 people (50-80 years): 21 - control group of volunteers, 15 - comparison group (patients with headache, degenerative-dystrophic changes in the spine, dyscirculatory encephalopathy of both sexes), 18 - patients with PD.
During the study, 8 autofluorescence spectra of the forearm skin surface were obtained from each patient using a spectrometer with excitation at a wavelength of 375 nm. The spectra were recorded in the range of 400-670 nm. Next 10 wavelengths that are most significant for discrimination were selected using step-by-step discriminant analysis. The obtained features were used for neural network classification into two groups (presence/absence of the disease). The classification accuracy on the test set was 98%, which is significantly higher than the accuracy obtained through linear discriminant analysis of 82%. The sensitivity and specificity were 96% and 99%, respectively.
Speaker
Nikita Pavlovich Bainaev-Mangilev
National Research Nuclear University MEPhI
Russia
Report
File with report
Discussion
Ask question
