Project Description Narrative:
Aphasia is a stroke-related language disorder that impacts around one in 250 people in Wisconsin and the U.S. This places aphasia's prevalence between Parkinson's and Alzheimer's disease, whilst impacting a younger population. The effects of aphasia are especially devastating in cases where stroke lesions damage the phonological system, located at the core of the middle cerebral artery territory. Here, people remain intellectually intact and able to mentalize the meaning of what they would like to say, but they cannot verbalize this as spoken, written, or typed words. As such, interactions with family, friends and colleagues are greatly limited, and victims often withdraw from their social roles and lose employment. This renders both victims and their close partners immediately vulnerable to ill-effects of social isolation, including depression, anxiety, ill health, and even early mortality. There is clearly an urgent need to develop methods to reinstate communication in aphasia. However, there is currently little hope of verbal recovery for people with large lesions to the phonological system with any known intervention. Despite decades of practice, speech and language therapy provides only modest benefit to extreme cases, thus radically different approaches are warranted.
Restoring communication would relieve the heavy societal health burden brought by social isolation. Developing remedial measures to reinstate communication in people with aphasia stands to impact Wisconsin at the societal level. In the first instance, restoring communication could help individuals retain and/or regain their position in society and continue their role in family and community matters and employment. This may help reverse the social and intellectual drain placed on society with each new case of aphasia. In the second instance, reinstating communication can reasonably be expected to relieve the social isolation faced by individuals with aphasia and their close partners. Given the heavy health burden associated with isolation (including a 50% increased risk of dementia, and a 60% chance of premature death), this benefit cannot be overstated.
Artificial intelligence (AI) could offer new hope by decoding messages from undamaged brain signals, but it needs "big data." This research team's long-term vision is to artificially restore verbal communication in individuals with aphasia. They seek to achieve this by developing brain- computer interfaces (BCI) to decode meaning from brain signals and transcribe this as words. Their immediate aim is to adapt AI models to transcribe brain signals elicited in the spontaneous conversations of people without aphasia.
Because modern-day AI methods are data hungry, this first step is essential to accumulate the big data set needed to introduce a proof-of-principle BCI before piloting aphasia applications.
MCW provides the ideal source to accumulate this data with a steady flow of patients undergoing 24/7 brain monitoring to guide epilepsy surgery (12 per year, potentially doubling by 2026). High-quality stereotactic electrophysiological brain signal recordings are maintained for approximately one week, during which patients engage in many recorded conversations with academics, visitors, and health staff. With patients' consent, these audio recordings can be experimentally transcribed with the assistance of speech transcription AI systems. With paired brain data and transcripts in hand, the research team will train AI models to transcribe continuous brain signals.
The project team will build brain decoding infrastructure by collating hundreds of hours of conversational brain data. The grand outcome of the proposed project will be an entry-level brain-to-text BCI framework that will provide a scalable foundation for a future aphasia oriented BCI. More specifically, besides the BCI, which on first pass can be expected to transcribe brain signals with modest accuracy, the team expect to accumulate approximately 300 hours of transcribed brain recordings (around 21 hours each for at least 24 participants over two years), as well as develop a bespoke software pipeline to support automated multi-talker audio transcription. This setup will be essential for future progression. The team anticipates that a fully operational brain-to-text BCI may require many thousands of hours of brain data to train based on the demands of speech recognition systems. Realistically, this is likely to require a multi-site data collection and transcription effort. By providing out-of-the-box transcription tools and a BCI that can be incrementally refined as new brain data becomes available, the team shall introduce a solution that can be parallelized across multiple epilepsy surgery sites with minimal within-site overheads. They contend that this ambitious endeavor may be the only prospect for re-instating communication in severe aphasia and thereby returning thousands of individuals to their roles in society.