Secondary Logo

Journal Logo

RFS – Blog

Artificial Intelligence and Applications in PM&R

Anderson, Dustin MD

Author Information
American Journal of Physical Medicine & Rehabilitation: November 2019 - Volume 98 - Issue 11 - p e128-e129
doi: 10.1097/PHM.0000000000001171
  • Free


Artificial intelligence (AI) has novel applications in the world of rehabilitation. This article will provide a brief overview of machine learning with examples of how the technology is advancing various fields in medicine, including PM&R.

Artificial intelligence is an iterative process that refers to the capacity of machines to perceive information, retain it as knowledge, and apply it toward adaptive behaviors in an environment. Artificial intelligence is an umbrella concept, with virtual (informatics) and physical (robotics) branches. Machine learning (ML) refers to specific methods for building algorithms that improve automatically with experience.

Artificial intelligence is ubiquitous in society, and Apple's Siri and Amazon's Alexa are competing to be your personal assistant. Google is using these tools to enhance spam filters. Tesla is building self-driving vehicles. Your credit card company uses ML to identify fraudulent purchases, and Netflix and Pandora uses it to determine what they will recommend.

Machine learning involves using known quantities from data to make predictions. Categories of ML generally include supervised learning, unsupervised learning, reinforcement learning, active learning, and semisupervised learning.

In supervised learning, the goal is to match inputs to outputs; thus, examples are labeled. When you identify spam in Gmail, that message is added as a labeled unit. Unsupervised learning allows machines to discover structure in the data, for example, clustering data based on different properties. Reinforcement learning uses rewards and punishments based on actions in a dynamic environment, as in self-driving cars. In active learning, the machine requests labels for individual examples, and semisupervised learning uses features of the first two.

Modeled from human neurons, deep learning uses layered artificial neural networks with weighted inputs and bias to determine solutions and has accelerated in use with increased computing power and data sources.

Advantages of machine learning include the ability to handle extremely large numbers of variables in complex environments. Machines require significant data points and dimensionality to outperform standard statistical analyses and can potentially “overfit” data with false correlations if multiple collinear and correlated variables are present. Machine learning does not solve the fundamental problem of causal inference in observational data.

In health care, AI achieved roughly 90% accuracy in identifying age-related macular degeneration,1 outperformed cardiovascular risk algorithms,2 and performed equal or better than the average dermatologist in detecting skin cancer.3,4

Positron emission tomography and magnetic resonance imaging scans were studied to predict cognitive decline, reducing sample estimates for a randomized controlled trial by a factor of five.5

In rehabilitation, ML is being used for symbiotic neuroprosthestics and myoelectric control,6 brain computer interface technology,7,8 perioperative medicine,9 and more. Machine learning methods have been used in musculoskeletal medicine such as diagnostic imaging, patient data measurement, and clinical decision support.10 In therapy, an artificial cognitive application was used to judge rehabilitative exercise based on the machine's indications.11

These ideas are rapidly being translated to our field and reinforced with experiential knowledge. As PM&R providers, we have the unique opportunity to channel these tools to advance care for our patients and the field at large.


1. Burlina PM, Joshi N, Pekala M, et al.: Automated grading of age-related macular degeneration from color fundus images using deep convolutional neural networks. JAMA Ophthalmol 2017;135:1170–6
2. Weng SF, Reps J, Kai J, et al.: Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLoS One 2017;12:e0174944
3. Esteva A, Kuprel B, Novoa RA, et al.: Dermatologist-level classification of skin cancer with deep neural networks. Nature 2017;542:115–8
4. Haenssle HA, Fink C, Schneiderbauer R, et al.: Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol 2018;29:1836–42
5. Ithapu VK, Singh V, Okonkwo OC, et al.: Imaging-based enrichment criteria using deep learning algorithms for efficient clinical trials in mild cognitive impairment. Alzheimers Dement 2015;11:1489–99
6. Sanchez JC, Mahmoudi B, DiGiovanna J, et al.: Exploiting co-adaptation for the design of symbiotic neuroprosthetic assistants. Neural Netw 2009;22:305–15
7. Pokorny C, Klobassa DS, Pichler G, et al.: The auditory P300-based single-switch brain-computer interface: paradigm transition from healthy subjects to minimally conscious patients. Artif Intell Med 2013;59:81–90
8. van Dokkum LE, Ward T, Laffont I: Brain computer interfaces for neurorehabilitation – its current status as a rehabilitation strategy post-stroke. Ann Phys Rehabil Med 2015;58:3–8
9. Patriarca R, Falegnami A, Bilotta F: Embracing simplexity: the role of artificial intelligence in peri-procedural medical safety. Expert Rev Med Devices 2019;16:77–9
10. Tack C: Artificial intelligence and machine learning | applications in musculoskeletal physiotherapy. Musculoskelet Sci Pract 2019;39:164–9
11. Simonov M, Delconte G: Humanoid assessing rehabilitative exercises. Methods Inf Med 2015;54:114–21

Artificial Intelligence; Machine Learning; Big Data; Rehabilitation; Physiatry; PM&R; Data Science

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.