In a landmark study, scientists from the BrainGate research collaborative have achieved a significant milestone that could transform the lives of individuals who have lost the ability to speak due to paralysis.
Published in the journal Nature, the research details how, for the first time, neural activity related to speech can be decoded into words on a screen, simply through the patient’s thoughts of speaking them. This development promises to open new horizons in the field of assistive technology, particularly for those suffering from conditions such as amyotrophic lateral sclerosis (ALS), also known as Lou Gehrig’s disease.
The centerpiece of this groundbreaking study is a patient with ALS, a condition that progressively weakens muscles and impacts physical function, including speech. Despite losing the ability to produce intelligible speech, this individual’s journey represents a beacon of hope.
By implanting sensors in specific areas of the cerebral cortex known to be involved in speech, the researchers could interpret the patient’s attempts at speaking movements, translating these into text with unprecedented accuracy, speed, and vocabulary breadth.
“This is a scientific proof of concept, not an actual device people can use in everyday life,” said Frank Willett, one of the study’s lead authors and a research scientist at Stanford University and the Howard Hughes Medical Institute. “It’s a big advance toward restoring rapid communication to people with paralysis who can’t speak.”
This research is a part of the BrainGate clinical trial, spearheaded by Leigh Hochberg, a critical care neurologist and professor affiliated with Brown University’s School of Engineering and the Carney Institute for Brain Science. Alongside Hochberg, Jaimie Henderson, a neurosurgery professor at Stanford University, and the late Krishna Shenoy, also from Stanford, have been pivotal in conducting this study.
The study involved positioning microelectrode arrays within the ventral premotor cortex and Broca’s area, regions of the brain associated with speech production and planning. Despite previous notions about Broca’s area, this study revealed that it holds minimal information regarding actual speech production, a finding that challenges longstanding assumptions in neuroscience.
Remarkably, the arrays placed in the ventral premotor cortex (area 6v) demonstrated an astonishing capability to differentiate between a variety of orofacial movements, phonemes (the smallest units of sound in speech), and words. The success rates here were compelling: 92% accuracy for orofacial movements, 62% for phonemes, and 94% for word recognition.
The real-time application of this research saw the patient, Pat Bennett (referred to as T12 in the study), engaging in attempts to speak sentences, which were then decoded using a recurrent neural network (RNN) and language models. This approach allowed for the real-time rendering of thoughts into text on a screen, a method that significantly surpassed the limitations of prior brain-computer interface (BCI) technologies in terms of vocabulary size and decoding speed.
One of the study’s highlights was the demonstration of the neural decoder’s ability to interpret sentences from a vast vocabulary at speeds mirroring normal conversation – an impressive 62 words per minute. This achievement is not just about the numbers; it’s a monumental step towards restoring the fundamental human experience of communication for those affected by severe speech impairments.
“For those who are nonverbal, this means they can stay connected to the bigger world, perhaps continue to work, maintain friends and family relationships,” Bennett wrote in an email.
“Imagine,” Bennett added, “how different conducting everyday activities like shopping, attending appointments, ordering food, going into a bank, talking on a phone, expressing love or appreciation — even arguing — will be when nonverbal people can communicate their thoughts in real time.”
However, the research team is cautious to note that while these results are promising, there’s a path ahead before this technology can be broadly applied in clinical settings. The challenges include refining the system to adapt to neural activity changes over time without extensive recalibration and further improving the decoding algorithms for even higher accuracy and usability.
The study, “A high-performance speech neuroprosthesis,” was authored by Francis R. Willett, Erin M. Kunz, Chaofei Fan, Donald T. Avansino, Guy H. Wilson, Eun Young Choi, Foram Kamdar, Matthew F. Glasser, Leigh R. Hochberg, Shaul Druckmann, Krishna V. Shenoy, and Jaimie M. Henderson.