Pályázati információ

2020-1.1.2-PIACI-KFI-2020-00011

 

Pain measuring device based on biological signal detection

József CSERNUS, Judit MÁRTON, János MÁRTON

Kereken-pálya Kft

Pain is a subjective feeling, which is influenced by an individual's emotional state, anatomy and physique. To date, however, a reliable evaluation for an objective assessment of pain intensity is not available. Therefore, a numerical or visual rating scale (usually 0-10) is the most up-to-date that is reported by the patient (Köny et al., 2012). These are one-dimensional assessment tools that can be considered effective in assessing acute pain, but require interactive communication between the patient and their physician and are therefore not suitable for non-communicative patients, or patients under anesthesia, or for patients who suffer from severe illness.

Kereken-Pálya Ltd. developed a pain measuring device and accompanying technology based on the measurement of skin resistance and other additional physiological parameters that is compact, easy-to-use, and which helps the doctor, nursing staff, researchers or the caregiver in the decision making regarding dosing, and in the evaluation of the efficiency of analgesics.

Many properties of the human body are based on electrical signals. These electrical signals are measurable and with their help we could draw conclusions. In the course of the project, we primarily developed a sensor technology that can effectively measure the various electrical properties of the skin. We created a modular hardware and software system that can provide evaluable information for professionals on the qualitative and quantitative determination of pain (sensation), that is based on advanced sensor technology and sensor fusion, uses machine learning solutions, has the appropriate interfaces to be integrated into other systems and is suitable for validation in a clinical environment.

The aim of our applied research was to map, analyze and synthesize the currently available scientific literature on pain measurement based on skin resistance measurement, which provided the basis for the creation of the measurement method of Kereken-Pálya Ltd.

In Jiang et al.’s 2018 study a method for continuous pain monitoring was developed using a variety of physiological parameters, heart rate (HR), respiratory rate (BR), galvanic skin reaction (GSR), facial surface electromyogram, and machine learning, which was classified based on the response to pain stimuli collected during heat and electrical treatment in 30 healthy volunteers. In the parameter matrix the The GSR, HR, and BR values correlated most with the level of pain intensity.

In their research Wang et al. (2020) used a hybrid RNN classifier to assess pain intensity. The RNN, which includes a bidirectional LSTM network was created for the abstract temporal representation of the pain data and furthermore they used a variety of physiological methods that are strong indicators of clinical pain based on their research findings.

Yamamoto et.al. (2006) aimed to determine the usefulness of measuring skin impedance for shoulder pain and compared their results with the VAS scale during the evaluation. They concluded that a combination of a subjective and an objective method, such as VAS, and measurement of skin impedance is suggested for the accurate diagnosis of the patient’s pain and appropriate treatment, which will lead to a more accurate assessment of the patient’s pain.

The aim of the study by Alvarez et al. (2019) was to examine changes in skin conductivity (SC) during eye examinations of ROP screening and to study their association with changes in heart rate and oxygen saturation. Premature babies who are at high risk of developing preterm retinopathy (ROP) need to have regular retinal examinations, which can be painful. Accurate measurement of pain in newborns remains a challenge, and more objective and reliable tools are needed. Their results show that NSFCs increased significantly during ROP screening eye examinations, coupled with frequent changes in heart rate and desaturation episodes. Therefore, it can be concluded that these changes may at least be partially related to pain, but this needs to be investigated by further research.

Based on the results of our scientific literature research, we designed our measurement methodology, created the development infrastructure, futhermore we amended our initial impedance measurement sensor set with additional pulse and blood oxygen measuring sensors as concluded in the scientific literature research, During the experimental development, we designed and manufactured the pain detection device prototypes, which are shaped as pulseoximeters. We performed validation tests of the pain detection device involving 42 volunteers, using artificial intelligence technology based on the classified responses collected during the treatment of pain stimuli caused by heat, ice, massage, and blood glucose spear.

We used standard machine learning methods (KNN and neural network) to teach the artificial intelligence.

A defining branch of artificial intelligence is machine learning. Machine learning systems are algorithms that are primarily based on statistics, which without defining any explicitly specifying rules are able to determine the control characteristics of a given process based on samples recorded from the process. There are several machine learning algorithms, such as linear and logistic regression, Hidden Markov Models (HMM), Support Vector Machines, decision tree, random forest, and gradient boosting-based techniques. These all use different approaches to data modelling (Gyires-Tóth, 2019).

Based on the results of our applied research and experimental development, we concluded that the pain signal generated by different methods can be separated using the K-Nearest Neighbors algorithm. This algorithm has proven to be extremely useful for analytical applications, furthermore “Random Forest” and similar regressors have also achieved relatively good accuracy. Coefficient analysis of the regressors also showed that although both sensory data and user attributes are involved in estimating pain levels, user attributes are embedded with much greater weight than other factors. One interpretation of this may be that the level of pain reported is a personal decision, therefore it is important to include user attributes when designing the final product.

The pain detection device system developed during the current research and development project of Kereken-Pálya Ltd is able to estimate the maximum pain level experienced by an individual based on the sensor data of impedance measurement, heart rate and blood oxygen level and from the individual’s attributes and lifestyle.

References

Marcus Köny, Michael Czaplik, Rolf Rossaint, Steffen Leonhardt (2012). Application of skin conductance measuring to assist pain therapy

Mingzhe Jiang, Riitta Mieronkoski, Elise Syrjälä, Arman Anzanpour, Virpi Terävä, Amir M. Rahmani,Sanna Salanterä, Riku Aantaa, Nora Hagelberg, Pasi Liljeberg (2018). Acute pain intensity monitoring with the classification of multiple physiological parameters, Journal of Clinical Monitoring and Computing, 26 June 2018

Bálint Gyires-Tóth (2019). A mélytanulás múltja, jelene és jövője, Budapesti Mûszaki és Gazdaságtudományi Egyetem, Távközlési és Médiainformatikai Tanszék, 2019

Run Wang, Ke Xu, Hui Feng, and Wei Chen (2020). Hybrid RNN-ANN Based Deep Physiological Network for Pain Recognition (2020)

Nobuyuki Yamamoto, MD; Eiji Itoi, MD; Hiroshi Minagawa, MD; Nobutoshi Seki, MD; Hidekazu Abe, MD; Yoichi Shimada, MD; Kyoji Okada, MD, (2006). Objective Evaluation of Shoulder Pain by Measuring Skin Impedance, Orthopedics, December, 2006 Volume 29

Alejandro Avila-Alvarez, Lorena Vazquez Gomez, Andrea Sucasas Alonso, Henar Romero Rey (2019). Skin conductance to assess pain and stress during retinopathy of prematurity screening, An Pediatr (Barc). 2020;92(6):365-375