AI Robotic Glove May Help Stroke Victims Play Piano Again

Geralt/Pixabay

Artificial intelligence (AI) combined with robotics is offering hope to those who have suffered neurotrauma. New research published in the journal Frontiers shows how an AI-powered soft robotic glove can help patients with neuromuscular disorders relearn how to play the piano.

“In the past, other soft robotic actuators have been used to play the piano; however, ours is the only one that has demonstrated the capability to ‘feel’ the difference between correct and incorrect versions of the same song,” wrote the researchers, affiliated with Florida Atlantic University, Boise State University, and the University of Florida College of Medicine.

Neurotrauma, injury to the brain and/or spinal cord, is a worldwide health problem. Each year, an estimated 69 million people globally sustain a traumatic brain injury (TBI), according to the World Health Organization (WHO). Annually around the world, an estimated 12.2 million new strokes occur, and one in four people over the age of 25 will have a stroke in their lifetime according to the World Stroke Organization (WSO). The WSO estimates that globally there are 101 million people who are living with stroke aftermath, a number that has nearly doubled in the last 30 years.

The researchers outfitted a soft robotic exoskeleton with piezoresistive sensor arrays with 16 taxels for five fingertips.

Piezoresistive sensors measure pressure with a high degree of accuracy. Their name reflects the piezoresistive effect, which is the change in electrical resistance when stress or strain is applied mechanically. Taxels (TActile piXEL) are sensors that can recognize contact pressure by calculating how much force is being applied to an area.

The researchers produced 10 song variations of “Mary Had a Little Lamb” (one correct and nine with rhythmic errors) and trained random forest (RF), K-nearest neighbor (KNN), and artificial neural network (ANN) algorithms with data collected from the five fingertip sensors.

A random forest algorithm, also known as a random decision forest, is a type of user-friendly machine learning algorithm that uses supervised machine learning. In artificial intelligence, supervised machine learning refers to a technique that uses labeled input data to train an algorithm to make predictions or classify data. These algorithms are widely used for classification and regression tasks. Instead of using just one decision tree, random forest algorithms consist of many individual decision trees that work together as an ensemble, hence the name “forest.” Class predictions from each decision tree are made and the one with the majority of votes becomes the final output.

K-nearest neighbor algorithms are also popular supervised machine learning algorithms that are commonly used for classification and regression problems. This basic algorithm stores all available cases instead of performing calculations, and then performs classification based on similarity. It is also considered a non-parametric method because the algorithm does not have assumptions about the data distribution. The algorithm looks at nearest annotated data point, or nearest neighbor, in order to classify a data point, hence the algorithm’s name.

Artificial neural networks, also known as neural nets or neural networks, are machine learning algorithms with architecture inspired by the biological brain. Artificial neural networks consists of an input layer, one or more hidden layers, and an output layer, where each layer contains interconnected multiple artificial neurons, called nodes, that have an associated weight and threshold.

Data is passed to the next layer of the network when a node is activated when it is above a specified threshold value. Artificial neural networks are able to rapidly classify data and are often used for search, voice recognition, and image recognition.

Deep learning, a subset of machine learning, is an artificial neural network with at least three layers. The more layers, the deeper the network. Machine learning (ML) is a subset of artificial intelligence where the algorithms are not hard-coded. Instead, in machine learning, the algorithm learns from training data in order to find patterns and make predictions.

The scientists harness the pattern-recognition capabilities of AI machine learning to “feel” the difference between right and wrong versions of a piano song. The team reported that the AI algorithm that produced the highest degree of accuracy was the artificial neural network algorithm, with more than 97% classification accuracy, plus or minus two percent with an able-bodied 25-year-old male playing. Without the human subject playing, the ANN algorithm on its own was able to perform classification with 94.6% accuracy, plus or minus 1.26 percent.

“These findings highlight the potential of the smart exoskeleton to aid disabled individuals in relearning dexterous tasks like playing musical instruments,” the scientists concluded.

Copyright © 2023 Cami Rosso All rights reserved.