K.S. Myshenkov1, Nekoula Haddad 2

1,2 Bauman Moscow State Technical University (Moscow, Russia)
 

In medicine, diagnosing diseases is a very difficult task. It relies on the accurate analysis of many clinical and pathological data of the patient by medical experts, which is a complex process.

Purpose: With the development of machine learning and information technology, researchers and practitioners are very interested in developing automated early disease prediction systems that are highly accurate, efficient, and useful for early diagnosis. This will minimize the risks of complications, improve the quality of treatment and avoid possible hospitalization. The article is devoted to the problem of predicting neurological diseases.

An analysis of the literature regarding the use of different machine learning methods for disease prediction was performed. Diagnosing Parkinson's disease was considered. Based on the analysis of the literature, the gradient boosting method was chosen as classifier to solve the problem. Assessing motor disorders features collected from 85 subjects with and without Parkinson’s disease were analyzed, these features are: typing speed, alternating finger tapping result, and single key tapping result. The obtained results are presented and the influence of every feature on the quality of classification is calculated.

Our studies have shown that the most important features in diagnosing Parkinson's disease are alternating finger tapping result, and single key tapping result, while typing speed turned out to be the less important feature. The results obtained are of interest to specialists involved in the prediction of morbidity and university students.

The article considers the problem of choosing a machine learning algorithm for predicting a neurological disease - Parkinson's disease. Scientific articles on neurological diseases and the use of artificial intelligence in diagnosing these diseases were analyzed, where, Alzheimer's disease receives the largest number of studies compared to Parkinson's disease and schizophrenia. In this work, the task was set to predict and diagnose Parkinson's disease using machine learning methods.

Based on the literature review, it can be concluded that the main methods used for disease prediction problems are: machine learning, deep learning, data analysis, and data mining. A comparative analysis of different machine learning algorithms explaining areas of their application was carried out, with an overview of the criteria that are taken into account when choosing an algorithm.

Based on the analysis performed, to solve the problem of predicting Parkinson's disease, the gradient boosting method was chosen using different sets of features, since this method shows good results for solving this class of problems. When solving the problem, the features of assessing motor disorders in Parkinson's disease studies of 85 subjects were analyzed. The following features were analyzed - information about the keystrokes by the subjects: typing speed, alternating finger tapping result, and single key tapping result. The influence of each feature on the quality of classification was calculated using the Shapley values equation. Our studies have shown that the best result in diagnosing Parkinson's disease was obtained when using the alternating finger tapping result, and single key tapping result features, while typing speed turned out to be the least important feature. At the next stage, the classification problem was carried out using each two features separately. Based on the results presented in the paper, it can be concluded that the best results in terms of the area under the curve and the duration of training were obtained when using the alternating finger tapping result, and single key tapping result features.

However, analyzing Parkinson's features extracted from uncontrolled typing can seem like a complex task, typing style varies greatly from person to person, and typed text varies a lot. Further, the difference in feature values between healthy and Parkinson's disease sufferers may not be related to the pathophysiology of Parkinson's disease and be mixed with volitional action or other diseases that may impair motor function. In the next works on disease prediction, it is planned to consider other methods of disease classification, data analysis, and feature engineering that ensure and improve disease prognosis. The results obtained may be of interest to professionals involved in predicting morbidity and university students.

Myshenkov K.S, Nekoula Haddad Neurological disease prediction using machine learning methods. Dynamics of complex systems. 2021. V. 16. № 1. P. 66−74. DOI: 10.18127/j19997493-202201-07 (In Russian)

1. Groves P. The big data revolution in healthcare: Accelerating value and innovation. McKinsey Quarterly. 2016.
2. Chen M. Big data: a survey. Mobile Networks and Applications. 2014. № 2. P. 171–209.
3. Jiang F. Artificial intelligence in healthcare: past, present and future. Stroke and Vascular Neurology. 2017. № 4. P. 230–243.
4. Jenni A.M. Machine learning in medicine: a practical introduction. BMC Medical Research Methodology. 2019. № 1. P. 1–18.
5. Sonawane J.S. Prediction of heart disease using multilayer perceptron neural network. International conference on information communication and embedded systems (ICICES2014). 2014. P. 1–6.
6. Keto J. Cardiovascular disease risk factors in relation to smoking behaviour and history: a population-based cohort study. Open Hear. 2016. № 2. P. e000358.
7. Jahmunah V. Automated detection of schizophrenia using nonlinear signal processing methods. Artif. Intell. Med. 2019. P. 101698.
8. Ortega-Zamorano F. Layer Multiplexing FPGA Implementation for Deep Back-Propagation Learning. Integr. Comput. Aided Eng. 2017. № 2. P. 171–185.
9. Raghavendra U. Artificial Intelligence Techniques for Automated Diagnosis of Neurological Disorders. European Neurology. 2019. № 1–3. P. 41–64.
10. Islam J. Brain mri analysis for alzheimer’s disease diagnosis using an ensemble system of deep convolutional neural networks. Brain Informatics. 2018. № 2. P. 1–14.
11. Fontana R. Early hippocampal hyperexcitability in ps2a pp mice: role of mutant ps2 and app. Neurobiol Aging. 2017. P. 64–76.
12. Mahmud M. Processing and analysis of multichannel extracellular neuronal signals: State-of-the-art and challenges. Front Neurosci. 2016. P. 248.
13. Hirtz D. How common are the «common» neurologic disorders? Neurology. 2007. № 5. P. 326–337.
14. Robin Harris J. Protein Aggregation and Fibrillogenesis in Cerebral and Systemic Amyloid Disease. 2012. P. 389.
15. Braak H. Invited Article: Nervous system pathology in sporadic Parkinson disease. Neurology. 2008. № 20. P. 1916–1925.
16. Harshvardhan T. Early Prediction of Parkinson Disease Using Machine Learning and Deep Learning Approaches. EasyChair Preprint. 2021. № 4889.
17. Geeta R. Handbook of Research on Disease Prediction through Data Analytics and Machine Learning. Manipal University Jaipur. 2020. № 1. P. 191–194.
18. Thube S., Rajeswari K. Analysis of Different Data Mining Tools using Classification, Clustering and Association Rule Mining. International Journal of Computer Applications. 2014. № 8. P. 35–39.
19. Usha R. D. Survey on Data Mining Tools and Techniques in Medical Field. International Journal of Advanced Networking & Applications. 2017. P. 51–54.
20. Krishnapriya S., Kalita D. Prediction of Heart Disease using Data Mining Techniques. Indian Journal of Science and Technology. 2016. № 39. P. 1–5.
21. Raudys S. Statistical and Neural Classifiers: An Integrated Approach to Design. Springer. 2001.
22. Summers M. J. Deep machine learning application to the detection of preclinical neurodegenerative diseases of aging. Scientific Journal on Digital Cultures. 2017. № 2. P. 9–24.
23. Ho T. K. Random decision forests perceptron training. ICDAR’95: Proceedings of the Third International Conference on Document Analysis and Recognition. 1995. P. 278–282.
24. Torsten H. RandomForest4Life: a random forest for predicting ALS disease progression. Amyotroph Lateral Scler Frontotemporal Degener. 2014. № 5–6. P. 444–452.
25. Debjani P. Predictive Systems: Role of Feature Selection in Prediction of Heart Disease. Journal of Physics Conference Series. 2019.
    № 1–6. P. 012074.
26. Sindhanaiselvan K. Forecasting Chronic Disease using Gradient Boosting Algorithm. International Journal of Engineering and Advanced Technology. 2019. P. 1066–1070.
27. Mathivanan G. An Ensemble Basedheart Disease Predictionusing Gradient Boosting Decision Tree. Turkish Journal of Computer and Mathematics Education. 2021. № 10. P. 3648–3660.
28. Giancardo L. Computer keyboard interaction as an indicator of early Parkinson's disease. Scientific Reports 6. 2016. № 1. P. 1–10.
29. Vsemirnaya organizaciya zdorov'ya. Vsemirnaya organizaciya zdravoohraneniya, 27 fevralya 2017 g. [Po sostoyaniyu na 12 dekabrya 2018 g.]. Dostupno po adresu: https://www.who.int/mediacentre/news/releases/2007/pr04/e/ (In Russian)