AI speech analysis project aims to improve brain injury care for veterans

A research project will use AI speech analysis to investigate communication difficulties in veterans with mild traumatic brain injury that standard assessments may miss.

The one-year study will examine whether speech patterns can provide objective evidence of cognitive and communication problems, with the aim of developing a tool for routine clinical use.

Researchers will analyse existing recordings of veterans describing their experiences to identify language patterns that could help clinicians assess and treat communication difficulties.

The project brings together researchers from The University of Texas at San Antonio and the South Texas Veterans Health Care System, supported by a US$200,000 grant from the San Antonio Medical Foundation.

It is being led by Rocio Norman, associate professor of communication sciences and disorders, Vivian Ta-Johnson, assistant professor of psychology, and Alicia Swan, director of rehabilitation research at the South Texas Veterans Health Care System.

Norman, a speech-language pathologist who has worked with veterans for 18 years, said many people with mild traumatic brain injury (mTBI) experience communication problems that conventional assessments fail to detect.

These include difficulty finding words, organising thoughts and following fast-moving or noisy conversations, which can affect employment, education, relationships and community life.

Norman said: “Many of the problems associated with mild TBI involve higher-level tasks.

“That differs from severe TBI, in which someone may need to relearn how to walk or live independently.

“With mild TBI, the difficulties may emerge when someone goes to college, returns to paid employment or tries to communicate meaningfully with loved ones.”

Many adult language assessments were developed for people with aphasia, a language disorder affecting communication, following a stroke, or for those with severe brain injuries.

Veterans with milder injuries may retain their vocabulary and grammar but struggle to organise ideas, tell a coherent story or respond to another person.

These skills depend on cognitive functions including attention, memory, planning and executive function.

Researchers hope speech could provide an objective measure of difficulties that are otherwise hard to identify.

Norman said: “The work is rooted in clinical experience with patients whose concerns could not be validated objectively.

“Objective measurement is fundamental to health care. We use the term ‘biomarker’ because we hope speech could become a biomarker for mild TBI.”

The project builds on an existing study led by Norman at the San Antonio Veterans Affairs healthcare system.

Researchers have already recorded veterans describing how they sustained their injuries and recounting an important life event, with recordings collected at several points during treatment.

The new study will use natural language processing (NLP), a form of artificial intelligence that analyses language, to examine transcripts of these recordings.

It will measure features such as how participants connect ideas, organise information, express emotion and provide detail.

The team will then compare these measurements with clinical information and other patient data.

Ta-Johnson, a social psychologist specialising in NLP and computational modelling, said: “Natural language processing gives us a way to turn patterns in language into measurable data. We can quantify different aspects of how people communicate, such as how they organise and connect ideas, and then examine those measures alongside clinical and other data.”

She added: “Recent technological advances have made some NLP tools powerful and accurate enough to capture and measure these facets of language in ways we could not before.”

Detailed language analysis currently requires time-consuming transcription and coding, often involving several reviewers.

An automated system could eventually reduce this workload and flag communication patterns for clinicians to consider alongside observations and standard assessments.

The researchers stress that the technology is intended to support clinicians, not replace them.

The study remains at the proof-of-concept stage.

Its immediate aim is to establish whether automated analysis can objectively identify communication differences and how they relate to symptoms such as anxiety, post-traumatic stress, depression and sleep difficulties.

It has not yet been established whether the approach can reliably identify these differences or provide clinically useful information during routine appointments.

If the pilot produces meaningful findings, the team plans to seek larger federal research grants and expand the work to broader patient populations, potentially across multiple sites.

Further studies could examine whether particular speech patterns reflect specific cognitive difficulties.

Norman hopes the research will eventually help clinicians recognise communication problems that affect veterans’ daily lives but may go undetected in standard assessments.

Objective evidence could also help veterans access appropriate care for difficulties affecting work, relationships, parenting and their return to community life.

Norman said: “It’s all for the patient and the betterment of patient care.”