|Year : 2022 | Volume
| Issue : 3 | Page : 117-119
Artificial neural network in clinical pain medicine and research
Prateek Arora1, Samarjit Dey2
1 Department of Anaesthesiology, All India Institute of Medical Sciences, Raipur, India
2 Department of Anaesthesiology, AIIMS, Guntur, Andhra Pradesh, India
|Date of Web Publication||21-Nov-2022|
Dr. Samarjit Dey
Department of Anaesthesiology, AIIMS, Mangalagiri, Guntur - 522 503, Andhra Pradesh
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
Arora P, Dey S. Artificial neural network in clinical pain medicine and research. Indian J Pain 2022;36:117-9
From the dawning of artificial intelligence (AI) in the field of medicine, many disputations have surfaced regarding machines taking over the role of doctors, its feasibility, loss of human touch, etc., AI has evolved its generation from machine learning to deep learning models, an example being artificial neural network (ANN), which, like the human brain, collects, processes, and yields data. Initial hurdles in yesteryear's ANN were challenging, and it was difficult to interpret the output data. Training with a large set of data that specifies input and desired responses, multilayered, backpropagation neural networks have refined the present-day AI to sync with the human understanding of cognitive science. This has paved the way for explainable AI. Machine learning imitates human learning, by obtaining data from input/“experiences,” and eventually becomes proficient to identify patterns and improve algorithms' accuracy. These newly acquired data are used to categorize, predict future information, and derive new knowledge. The predicted output is then compared to the actual output to realize the error in prediction, and feedback to the network is sent via the backpropagation channel to adjust the relevant weights [Figure 1].
|Figure 1: Schematic representation of ANN showing algorithm for pain assessment using facial cues and EEG. ANN: Artificial neural network, EEG: Electroencephalogram|
Click here to view
AI has found its niche in anesthesia, especially in trauma, perioperative period, drug dose monitoring, American Society of Anesthesiologists Physical Status (ASA-PS) score, and critical care. Likewise, machine learning models using the neural network are being used to train and provide continuous monitoring and mitigating postsurgical pain timely in neonates. Reasonably so models based on pain assessment tools in special populations, such as the elderly, small children, patients on mechanical ventilation, and patients of unsound minds, are particularly important for adding objectivity to pain description. ANN can integrate data obtained from pain assessment tools such as Wong-Baker FACES Scale, and Numerical Rating Scale, with that obtained from multipara vital monitors such as heart rate variability, rate and depth of respiration, blood pressure, pupillometry, etc. This co-relation can guide perioperative pain therapy. Like the tracts in the human brain, the deep neural network receives input from various sources, then modifies the weight of each input based on past experiences, and then summarizes a value to the output.
The pathophysiology of chronic pain is far more complex and multidimensional as compared to acute perioperative pain. The concept of total pain by Dame Cicely Saunders encompasses a person's physical, psychological, social, spiritual, and practical struggles. Common clinical conditions seen in a pain medicine outpatient unit include low backaches, neck pain, compressive neuropathies, cancer pain, fibromyalgia, headache, and facial pain. Considerable overlap exists between nociceptive, neuropathic, and nociplastic types of pain in one or more of the above conditions. Prolonged nociceptive inputs can further trigger the excitability of central nociceptive pathways, the phenomenon of central sensitization. Neuropathic pain assessment is difficult in routine clinical practice and relies on patients' descriptions of symptoms and questionnaires such as painDETECT, and The Leeds Assessment of Neuropathic Symptoms and Signs (LANSS). For instance, an ANN-based tool provided with inputs of symptoms of low backache, aggravating factors, pain map self-reported by the patients, added with clinical assessment data by the treating physician, and radiological imaging (magnetic resonance imaging) can help screen patients of low backache. With the help of machine learning, red flags can be evaluated, and the need for intervention and/or surgery can be objectively determined. Similarly, future prospects exist for wearable technologies for patients with headaches, such as migraine, to design individualized therapies based on the frequency of headache episodes, trigger factors, and rescue analgesia plans.
Pain is a manifestation of neural activity, the quest for a “pain biomarker” prevails. It can be “a machine, a system or a process capable of pattern recognition of neural activity in the brain correlated with pain.” Electroencephalography (EEG)-based machine learning models using support vector machine (SVM) algorithms have shown promising results in predicting “no-pain” versus “high-pain” states. EEG, even as wearable technology, has also been shown to accurately and significantly discriminate painful and healthy states in adolescents with chronic musculoskeletal pain, using two models of machine learning, namely, SVM and logistic regression. Systematic and regularly timed data input by patients into the ANN model can help develop tools and technologies to combat chronic pain states. AI-driven cognitive behavioral therapy for chronic pain-based mobile health monitoring has been shown to improve patient outcomes while more effectively targeting scarce clinical resources. Systematic reviews, furnish a glimmer of hope for prospects of AI in clinical chronic pain medicine; however, evidence in terms of which models of AI are best suited for a said clinical entity is still scarce.
Given the enormous time frames and data sets in research based on chronic pain, a need for a computer-based model for acquiring and processing complex information seems apparent. In a topical review, Lötsch and Ultsch have classified the machine learning tools in pain research into (a) for classification tasks, (b) for data structure detection, and (c) for knowledge discovery in experimental or clinical data. Among the tools available, regression models, convolutional and ANNs, and SVMs are the most frequently used. Amid the “pain-” based research utilizing AI/machine learning tools, no matches were found on a PubMed search from the Indian subcontinent. It may be prudent to state AI's poor penetrance in pain research in our region.
AI is to be seen as a contrivance, an extension of our clinical skills and assessment in pain medicine. An understanding of what AI is capable of without compromising the vulnerabilities in patient care is imperative in incorporating the technology into clinical practice. Ethical ramifications of the technology that we have created remain pertinent in our society. Moreover, AI is far from automating certain aspects of our job, rather they should be viewed as tools which enhance our capabilities as pain physicians and researchers.
| References|| |
Madanu R, Abbod MF, Hsiao FJ, Chen WT, Shieh JS. Explainable AI (XAI) applied in machine learning for pain modeling: A review. Technologies 2022;10:74.
Matsangidou M, Liampas A, Pittara M, Pattichi CS, Zis P. Machine learning in pain medicine: An Up-To-Date systematic review. Pain Ther 2021;10:1067-84.
Salekin MS, Mouton PR, Zamzmi G, Patel R, Goldgof D, Kneusel M, et al.
Future roles of artificial intelligence in early pain management of newborns. Paediatr Neonatal Pain 2021;3:134-45.
Ong CK, Forbes D. Embracing Cicely Saunders's concept of total pain. BMJ 2005;331:576.
Levitt J, Saab CY. What does a pain 'biomarker' mean and can a machine be taught to measure pain? Neurosci Lett 2019;702:40-3.
Levitt J, Edhi MM, Thorpe RV, Leung JW, Michishita M, Koyama S, et al.
Pain phenotypes classified by machine learning using electroencephalography features. Neuroimage 2020;223:117256.
Teel EF, Ocay DD, Blain-Moraes S, Ferland CE. Accurate classification of pain experiences using wearable electroencephalography in adolescents with and without chronic musculoskeletal pain. Front Pain Res (Lausanne) 2022;3:991793.
Piette JD, Newman S, Krein SL, Marinec N, Chen J, Williams DA, et al
. Artificial Intelligence (AI) to improve chronic pain care: Evidence of AI learning. Intell Med 2022;6:100064.
Tagliaferri SD, Angelova M, Zhao X, Owen PJ, Miller CT, Wilkin T, et al.
Artificial intelligence to improve back pain outcomes and lessons learnt from clinical classification approaches: Three systematic reviews. NPJ Digit Med 2020;3:93.
Lötsch J, Ultsch A. Machine learning in pain research. Pain 2018;159:623-30.