The information is prepared on the basis of data from the information-analytical system RSF, informative part is represented in the author's edition. All rights belong to the authors, the use or reprinting of materials is permitted only with the prior consent of the authors.

Project titleDevelopment of high-sensitive methods for early and differential diagnosis of neurodegenerative diseases with trembling hyperkinesis

AffiliationKotelnikov Institute of Radioengineering and Electronics of Russian Academy of Sciences,

Research area 05 - FUNDAMENTAL RESEARCH IN MEDICINE, 05-702 - Medical informatics

KeywordsOptimization algorithms, neural network, regression trees, differential diagnostic, wave train electrical activity, time-frequency analysis, wavelet analysis, wave packets, signal processing, trembling hyperkinesis, EEG, EMG, Parkinson's disease, essential tremor, atypical parkinsonism

Annotation
Diagnosis of atypical parkinsonism (AP) is a very actual problem nowadays due to the growing number of elderly people in Russia and, as a consequence, an increase in the prevalence of the disease. Atypical Parkinsonism is a term used to refer to Parkinson's syndrome, which differs in its characteristics from Parkinson's disease (PD). The complexity of the diagnosis of AP is in that AP exhibits symptoms that are not characteristic of PD. For example, levodopa preparations have no effect, there is no rest tremor, a bilateral symmetrical beginning of disease is observed, etc. Due to the above mentioned difficulties, it is of a great importance to develop methods for automatic diagnosis of AP, PD, and essential tremor (ET) based on time-frequency analysis of biomedical signals, such as an electroencephalogram (EEG), electromyogram (EMG), and accelerometer signals. A promising method for solving this task is application of the wave-train-based analysis of electrophysiological signals. A wave train electrical activity is an increase of the power spectral density of the signal that is localized in space and time. The authors of the project use the parameters of local maxima in wavelet spectrograms for description of wave train biomedical electrical activity. Earlier, the authors of the project have demonstrated that statistical analysis of some features of EEG wave trains allows detection of PD at an early stage. Therefore, the methods for obtaining new quantitative features of AP, PD, and ET will be explored and developed within the framework of the project. The proposed methods are based on the evaluation of various characteristics describing wave trains of EMG, accelerometer, and EEG signals (namely, the central frequency of the wave trains, the duration of the wave trains at half maximum, the wave train bandwidth at half maximum, the average number of wave trains per second, etc.) and further statistical analysis of these characteristics. The developed methods will solve the actual problem of distinguishing the early symptoms of AP, PD, and ET. The developed methods and approaches will be tested on clinical data of electrophysiological diagnostics of AP, PD, and ET, which will be acquired for the project together FSBI Research Center of Neurology.

Expected results
1. Specialized optimization algorithms for time-frequency analysis of wave train electrical activity in biomedical signals for the differential diagnosis of patients with atypical parkinsonism, Parkinson's disease, and essential tremor. Objective functions of optimization algorithms, classification metrics of the studied data for the diagnosis of AP, PD, and ET. 2. Recognition methods for AP, PD, and ET based on separating hyperplanes, regression trees (decision trees) and neural networks for time-frequency analysis of wave train electrical activity in biomedical signals for differential diagnosis of patients with atypical parkinsonism, Parkinson's disease, and essential tremor. Neural network algorithms for automatic analysis and diagnosis of AP, PD, and ET at an early stage. Results of comparison of classical recognition methods based on dividing hyperplanes, decision trees with neural network methods. 3. Models of patients with AP, PD, and ET in the feature space of wave train characteristics of biomedical signals (EEG, EMG, accelerometer). 4. New quantitative features of atypical parkinsonism, Parkinson's disease, and essential tremor in EMG signals, mechanical tremor, and EEG, which will be obtained using the method of analysis of wave train electrical activity and visualization of the results of statistical analysis developed within the project. In the scope of this project, the search for the ranges of wave train parameters in the AUC value diagrams will be performed, in which the differences between the patient group and the control group are observed. Evaluation of the informativeness of the obtained new features of atypical parkinsonism, Parkinson's disease, and essential tremor using nonparametric statistics. 5. The results of a comparative analysis of the obtained features of atypical parkinsonism, Parkinson's disease, and essential tremor at an early stage and the assessment of the specificity of the obtained features for Parkinson's disease. The results of a quantitative comparison of new developed methods with existing methods, including a comparison of the effectiveness of these methods. 6. Experimental Matlab-based and Python-based software for detecting features of atypical parkinsonism, Parkinson's disease, and essential tremor based on the study of wave train parameters on wavelet spectrograms of EMG, tremor, and EEG signals. The system will search for regularities in the wave train electrical activity data. For this, optimization algorithms and neural networks will be used, which will investigate the combinations of features of AP, PD, and ET. The features of AP, PD, and ET are the number of wave trains, the parameters of which belong to certain ranges, calculated using optimization algorithms. Optimization algorithms will identify ranges of wave train parameters that are quasi-optimal in terms of distinguishing between patients and healthy subjects. Neural networks will be used to diagnose AP, PD, and ET based on the ranges of wave train parameters selected by optimization algorithms. The targeted system should implement the following functions: importing data in edf and txt formats, plotting EMG, tremor and EEG signals, calculating local extrema on wavelet spectrograms that meet the specified conditions, selecting and studying the statistical characteristics of wave trains on wavelet spectrograms (central frequency, duration, bandwidth, average number of wave trains per second, etc.), plotting AUC values versus characteristics of range boundaries (center frequency, duration, bandwidth, average the number of wave trains per second, etc.), searching with the help of optimization algorithms for subspaces in the multidimensional space of characteristics of wave trains on wavelet spectrograms, in which the patient group and the control group are well distinguished, checking the statistical significance of differences between groups of patients and control (using the Mann-Whitney test, taking into account the Bonferroni correction). The experimental software will be used to build models of patients with AP, PD, and ET in the feature space of characteristics of wave trains of biomedical signals (EEG, EMG, accelerometer) for differential diagnosis. 7. The results of clinical approbation of the software created within the framework of the project. Results of comparison of the method of analysis of wave train electrical activity with standard clinical methods for diagnosing AP, PD, and ET. The results of the analysis of wave train electrical activity carried out during the treatment of patients with AP, PD, and ET at the FSBI Research Center of Neurology. For this, experimental data from joint electrophysiological and mechanical studies of at least 10 patients with atypical parkinsonism, at least 10 patients with the first stage of Parkinson's disease, and at least 10 patients with essential tremor and at least 10 volunteers of the control group will be collected. The scientific significance of the expected results is that a new exploratory analysis method will be developed that can be applied to analyze other types of biomedical signals. The social significance of the expected results of the project lies in the fact that the use of computer diagnostic methods of AP, PD, ET will help reduce the burden on the primary care of neurologists, help to further improve the quality of diagnostics and reduce social stress associated with insufficient resources and medical personnel to provide timely and quality care for patients with symptoms of neurodegenerative diseases. The developed methods and algorithms will allow solving the urgent problem of differential diagnosis of atypical parkinsonism, Parkinson's disease, and essential tremor in the early stages of the disease. The project is being carried out jointly with the Moscow organization of the Research Center of Neurology, in which the results of the project will be implemented for the diagnosis of patients with AP, PD, ET, and other socially significant diseases. The expected results are new and original.

Annotation of the results obtained in 2022
A novel approach to exploratory data analysis has been created, which specifically targets the identification of regular patterns in paired biomedical signals. An example of paired biomedical signals is a pair of electromyograms (EMG) in antagonist muscles (flexors and extensors in the patient's arm). The method enables the identification of the frequency ranges in which the instantaneous phase of the studied signals demonstrates neurophysiological regularities with diagnostic potential. Moreover, it allows for the determination of the optimal balance between the influence of signal phase and amplitude characteristics when searching for neurophysiological patterns. To address the challenge of identifying frequency ranges in EMG that exhibit neurophysiological regularities, the project team designed specialized types of AUC diagrams for time-frequency analysis of wave train electrical activity (AUC is the area under the ROC curve). Two new types of AUC diagrams have been developed. The first type of AUC diagrams is based on the use of the Hilbert transform to calculate the instantaneous phase of EMG signals. It enables the identification of frequency ranges where there exist differences in the values of the instantaneous phase of EMG signals among patients with various neurophysiological conditions. The second type of AUC diagrams is based on the use of cross-wavelet spectra of signals. It allows the identification of the frequency ranges where the phase regularities of the signals are present while considering the dominant contribution of EMG fragments with high amplitude. The project team also developed supplementary features that depict the relationship between paired EMG signals. Moreover, they introduced the concept of "cross-wave train" to describe this relationship. The cross-wave train is a time- and frequency-localized increase in the cross-wavelet spectrum. The characteristics of a cross-wave train are center frequency, power spectral density (PSD), duration in seconds and periods, bandwidth, and instantaneous phase. New neurophysiological regularities have been revealed in EMG signals of antagonist muscles in patients with Parkinson's disease (PD) and essential tremor (ET). The identified regularities exhibit opposite directions, shedding light on certain inconsistencies reported in the neurophysiological literature. The project team has designed dedicated optimization algorithms to enhance time-frequency analysis of wave train electrical activity in biomedical signals for differential diagnostics of atypical parkinsonism (AP), PD, and ET. With the use of the developed algorithms, it becomes feasible to automate the iterative process of refinement of the cross-wave train characteristics for specific neurodegenerative diseases. The study uncovered the presence of at least four types (frequency ranges) of cross-wave trains that discriminate between patients with PD and those with ET. These four types of cross-wave trains differ in the center frequency as well as in amplitude, duration, bandwidth, and instantaneous phase. Two of the four types of cross-wave trains are characterized by AUC values close to 1. Following the meaning of the AUC diagrams, AUC values close to 1 mean that cross-wave trains of these types occur in patients with PD but are not typical for patients with ET. The other two types of cross-wave trains are characterized by AUC values close to 0; these types of cross-wave trains are common in ET patients but are rare in PD patients. The project team conducted a correlation analysis of the count of cross-wave trains that correspond to various frequency ranges. The results of the correlation analysis indicate a statistically significant correlation between the number of cross-wave trains in the examined frequency ranges among patients with Parkinson's disease (PD). The identified regularities could serve as the basis for creating highly sensitive methods for distinguishing between PD and ET through diagnosis. Notably, the developed method is versatile and can be utilized for analyzing other types of biomedical signals as well. The results were published in January 2023 in the Sensors journal. The project team applied for a patent for the invention "A method for the differential diagnosis of essential tremor and the first stage of Parkinson's disease using the analysis of wave trains on the cross-wavelet spectrum of electromyographic signals of antagonist muscles". The theoretical basis of the method for analyzing wave train electrical activity, developed by the project executors for the study of biomedical signals, is presented in an article in the RENSIT journal. The concept of wave train electrical activity is considered, along with mathematical tools employed to examine the characteristics of wave trains observed in biomedical signals: histograms of wave train parameters and AUC diagrams. The findings indicate that the approach is not limited to the differential diagnosis of Parkinson's disease and essential tremor but can also be used to analyze epileptiform electrical activity. This work was carried out using electroencephalographic (EEG) data from laboratory animals (rats); EEG data were kindly provided by the Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences. Following the analysis of wave train electrical activity in the brains of rats, the study revealed that: (1) At least two classes of wave trains are observed in the composition of epileptic discharges, differing in the central frequency and bandwidth. The detected classes of wave trains have characteristic ranges of the instantaneous phase; this indicates that these wave trains have different signal shapes. (2) The background EEG contains two classes of wave trains differing in center frequency and bandwidth. The obtained results on peak-wave discharges of typical absence epilepsy observed in WAG/Rij rats demonstrate the potential of the developed signal analysis method to examine various forms of epileptic activity.

Publications

1. Sushkova O.S., Morozov A.A., Kershner I.A., Khokhlova M.N., Gabova A.V., Karabanov A.V., Chigaleichick L.A., Illarioshkin S.N. Investigation of Phase Shifts using AUC Diagrams: an Application to the Differential Diagnosis of Parkinson's Disease and Essential Tremor MDPI, Sensors, V. 23, Issue 3, P. 1531 (year - 2023) https://doi.org/10.3390/s23031531

2. Sushkova O.S., Morozov A.A., Petrova N.G., Khokhlova M.N., Gabova A.V., Karabanov A.V., Chigaleychik L.A., Sarkisova K.Y. Method of wave train electrical activity analysis – the theoretical basis and application RENSIT: Radioelectronics. Nanosystems. Information technologies, V. 14, No. 3, P. 317-330. (year - 2022) https://doi.org/10.17725/rensit.2022.14.317

3. Sushkova O.S., Morozov A.A., Gabova A.V., Chigaleychik L.A., Karabanov A.V. Исследование нейрофизиологических закономерностей болезни Паркинсона на первой стадии с помощью метода анализа всплескообразной электрической активности мышц Тезисы докладов 14-й Международной конференции Интеллектуализация обработки информации, стр. 423-424 (year - 2022)

4. - Способ быстрой дифференциальной диагностики эссенциального тремора и первой стадии болезни Паркинсона с помощью анализа всплесков на кросс-вейвлет спектре электромиографических сигналов мышц-антагонистов -, - (year - )

Annotation of the results obtained in 2023
The possibility of creating specialized neural network architectures suitable for training on small samples of EMG data in patients with neurodegenerative diseases was demonstrated. Computational experiments with clinical data confirmed the presence of temporal regularities in electromyograms (EMG) of patients with Parkinson’s disease (PD) and essential tremor (ET). A method for identifying features of freezing of gait (FOG) in Parkinson’s disease (PD) was developed. The method is based on the analysis of accelerometric signals. New neurophysiological regularities in the accelerometric signals of patients with PD were identified. These patterns are promising for recognizing cases of FOG in patients with PD. Methods for automatically searching for features of neurodegenerative diseases Parkinson’s disease and essential tremo based on AUC diagrams and optimization algorithms have been studied and developed. Namely, methods for automatically searching for global extrema in the multidimensional space of parameters of wave train electrical activity were investigated and developed. During experiments with optimization algorithms, a new feature of wave train electrical activity was discovered in the 1-4 Hz frequency range in EMG signals, which makes it possible to solve the problem of differential diagnosis of PD and essential tremor (ET) with an accuracy close to 100%. The presence of such feature in the specified frequency range is a new result in the field of neurophysiology. The problem of robustness of solutions of the optimization problem was studied. Two different methods were proposed for assessing the robustness of solutions of the optimization problem, calculated during the analysis of the space of wave train parameters. The first method is based on calculating the radius of stability of the solution under study. The second method is based on the physical meaning of the wave train parameter space. A comparison of stability estimates for solutions to the optimization problem based on the two indicated principles was made. Joint electroencephalographic, electromyographic, and accelerometric studies were carried out on 28 patients, namely, patients with atypical parkinsonism (AP) (3 patients), patients in whom ET evolves to PD (7 patients), patients with functional tremor (1 patient), patients in the 1st stage PD (13 patients), patients in the 2nd stage 2 PD (1 patient), patients with ET (3 patients), as well as 5 control group subjects. A comparison of two different methods for the differential diagnosis of PD and ET, proposed as part of the project, was made. Estimates of the accuracy, sensitivity, and specificity of these methods were obtained. The first method is based on the analysis of local maxima in wavelet spectrograms of accelerometer signals and envelopes of EMG signals. The second method is based on the analysis of local maxima in the cross-wavelet spectra of the envelopes of EMG signals of paired antagonist muscles. The second method was patented during the reporting period of the project. A comparison of the two indicated methods for the differential diagnosis of PD and ET based on two years of experience in the clinical application of these methods at the Research Center of Neurology showed that the method based on studying the characteristics (including phase) of cross-wave trains on a cross-wavelet-spectra of the EMG envelopes of the antagonist muscles of the patient’s limbs supplies better results than the method based on the analysis of the characteristics of wave trains in the wavelet spectrograms of the accelerometer signals and the envelopes of the EMG signals. It is shown that both methods are of interest from the point of view of identifying features of AP.

Publications

1. Gurgenidze A.V., Sushkova O.S., Khokhlova M.N., Morozov A.A. Investigating Parkinson’s Disease Gait Freezing Through the Analysis of Wave Train Electrical Activity Patterns IEEE Proceedings, - (year - 2024)

2. Sushkova O.S., Morozov A.A., Gabova A.V., Khokhlova M.N., Kershner I.A., Chigaleichick L.A., Poleschuk V.V., Karabanov A.V. Comparative Analysis of Methods for Differential Diagnosis of Parkinson's Disease and Essential Tremor Based on the Study of Wave Train Electrical Activity IEEE Xplore Digital Library, P. 1-6 (year - 2024) https://doi.org/10.1109/DSPA60853.2024.10510032

3. Sushkova O.S., Morozov A.A., Kershner I.A., Khokhlova M.N., Gabova A.V., Chigaleichick L.A., Karabanov A.V. Research and Development of Automatic Methods for Detecting Features of Parkinson’s Disease and Essential Tremor Based on AUC Diagrams IEEE Proceedings, - (year - 2024)

4. Sushkova O.S., Morozov A.A., Khokhlova M.N., Kershner I.A., Gabova A.V., Chigaleychik L.A., Karabanov A.V. Investigation and development of methods for automatic search for AUC-diagram-based features of Parkinson's disease and essential tremor RENSIT: Radioelectronics. Nanosystems. Information technologies, V. 16, No. 1, P. 67-78 (year - 2024) https://doi.org/10.17725/j.rensit.2024.16.067

5. Gurgenidze A.V., Sushkova O.S., Khokhlova M.N., Morozov A.A. Исследование замораживания походки при болезни Паркинсона методом анализа всплескообразной электрической активности Издательство Самарского университета, - (year - 2024)

6. Sushkova O.S., Morozov A.A., Kershner I.A., Khokhlova M.N., Gabova A.V., Chigaleichick L.A., Karabanov A.V. Исследование и разработка методов автоматического поиска признаков болезни Паркинсона и эссенциального тремора на основе AUC-диаграмм Издательство Самарского университета, - (year - 2024)

7. Sushkova O.S., Morozov A.A., Kershner I.A., Khokhlova M.N., Gabova A.V., Karabanov A.V., Chigaleichick L.A., Illarioshkin S.N. Разработка и применение метода обнаружения фазовых сдвигов в парных биомедицинских сигналах для дифференциальной диагностики болезни Паркинсона и эссенциального тремора Математические методы распознавания образов: Тезисы докладов 21-й Всероссийской конференции с международным участием «Математические методы распознавания образов» (ММРО-2023), г. Москва-2023, С. 232 (year - 2023)

8. Sushkova O.S., Morozov A.A., Gabova A.V., Karabanov A.V., Chigaleychik L.A., Illarioshkin S.N. Способ дифференциальной диагностики эссенциального тремора и первой стадии болезни Паркинсона с помощью анализа всплесков на кросс-вейвлет спектре электромиографических сигналов мышц-антагонистов Патент номер RU 2797878 C1., Патент номер RU 2797878 C1. Опубликовано: 09.06.2023 Бюл. № 16. – Заявка на патент номер RU 2022122588, 22.08.2022. – Дата начала отсчёта срока действия патента: 22.08.2022. – Дата регистрации: 09.06.2023. (year - 2023)