Methodology for assessing and predicting personalized occupational health risks based on adaptive neural fuzzy network for image recognition
N.V. Zaitseva1,2, А.А. Savochkina3, М.А. Zemlyanova1,3, D.V. Goryaev4, А.G. Fadeev4
1Federal Scientific Center for Medical and Preventive Health Risk Management Technologies,
82 Monastyrskaya St., Perm, 614045, Russian Federation
2Russian Academy of Sciences, Department for Medical Sciences, 14 Solyanka St., Moscow, 109240,
Russian Federation
3Perm National Research Polytechnic University, 29 Komsomolskii Av., Perm, 614990, Russian Federation
4Federal Service for Surveillance over Consumer Rights Protection and Human Wellbeing, Krasnoyarsk Regional Office, 21 Karatanova St., Krasnoyarsk, 660097, Russian Federation
Health protection provided for industrial production workers is a national priority, which determines possibilities for preservation of occupational longevity. Given that, it is becoming especially vital to create and develop scientific grounds for analyzing occupational health risks associated with complex exposures to occupational and work-related risk factors with special emphasis placed on personalized estimates. In this study, we aimed to develop and test a methodology and software for assessing and predicting personalized occupational health risks based on an adaptive neural fuzzy network for image recognition.
The study design was based on an artificial intellect model as a mathematical structure trained to recognize regularities and establish whether an analyzed object belonged to a specific occupational health risk category per a system of indicators. Network training and validation were performed on an example sample made of workers employed at underground copper-nickel ore mining using data on their working conditions, exposure factors and individual biomedical indicators (175,000 parameters overall). The training sample equaled 80 % and the validating one 20 %. The network was tested on an independent sample of data about workers exemplified by blast-hole drillers as a basic occupation at the mine.
A methodology was developed and provided with relevant software; its theoretical ground was represented by an adaptive neural fuzzy network for image recognition. The network had a specific hybrid multilayer architecture, which ensured accuracy of predictive estimates and error minimization. Personalized occupational health risks for each worker in the validating sample were caused by vibrational disease associated with simultaneous exposure to occupational noise
(10–40 dBA higher than MPL) and total vibration equal to 106–113 dB; these risks were ranked as ‘high’ and ‘very high’. Health risks caused by sensorineural hearing loss (SHL) associated with combined exposure to noise (5–30 dBA higher than MPL) and adverse chemicals (2.0–2.5 times higher than single maximum MPC) were estimated as varying from medium to very high. Health prediction for workers of this occupation in the independent sample showed that vibration diseases accounted for 75 % of expected occupational and work-related diseases with risks varying form low to high; polyneuropathy, 48 %; SHL, 6 %; dorsopathy, 75 %; essential hypertension, 30 %. Profound medical examination of blast-hole drillers confirmed that these health risks were actually manifested as diseases in 87–89 % workers.
The developed and tested methodology is quite effective. The prediction accuracy is estimated to reach 89 ± 2 % and the prediction error trend comes to minimum. The methodology provides a considerably wider opportunity to obtain prompt and accurate personalized prediction of health risks for workers. The system is eligible for workers employed at variable productions and implements a transition from contact-based examinations to quantitative prediction without any information losses, which determines its scalability and possibility to replicate it.