Leveraging AI to help stroke survivors recover speech abilities
Doctoral student Shreya Parchure and her team evaluated the usefulness of an AI tool for personalizing speech therapy for patients with post-stroke aphasia.
4 min. read
Shreya Parchure, an M.D.-Ph.D. candidate at Penn, conducts much of her AI-driven research in the Laboratory for Cognition and Neural Stimulation, focusing on ways to personalize speech therapy for patients with post-stroke aphasia.
During clinical rotations in a neurocritical care unit, Shreya Parchure’s interactions with patients solidified her desire to advance research and care for those with post-stroke aphasia. This impaired ability to understand or produce speech affects one-third of stroke survivors and can cause long-term language deficits. One patient, Parchure recalls, was initially unable to speak, but gradually, through speech therapy, began regaining words day by day.
Speech therapies for post-stroke aphasia are typically standardized. In a recent study, however, Parchure and her team in the Laboratory for Cognition and Neural Stimulation (LCNS) explored whether “explainable AI”—a set of machine learning methods and approaches focused on providing rationale behind results, enabling human users to better interpret and trust recommendations—could help doctors tailor treatment by predicting the most effective path to language recovery.
Some AI models have looked at neuroimaging and length of time from a stroke to determine the severity of aphasia, but Parchure and colleagues expanded these methods by accounting for how language is formed in and processed by the brain. Explainable AI, Parchure says, can integrate clinically available data—such as age, education, or the size of a stroke—with the linguistic difficulty of words. This approach enables the AI model to help predict recovery time and suggest treatments to physicians. The AI model also provides a clinical rationale for those treatments based on the patient’s unique situation.
“When we have an AI making a prediction, we really want to know why,” says Parchure, who also earned her bachelor’s and master’s degrees in bioengineering from Penn.
Exploring the utility of explainable AI
Mentored by Roy H. Hamilton, director of LCNS and professor of neurology, physical medicine and rehabilitation, and psychiatry in the School of Medicine, Parchure collected speech samples from patients with post-stroke aphasia as part of a team. She used this data to train an explainable AI algorithm, testing it to account for various language tasks and make predictions for patient recovery based on a diverse range of clinically relevant information.
Parchure and her colleagues found that the AI model made detailed evaluations of patients’ experiences with language production beyond the usual information provided through intake forms and clinical testing scores, probing deeper for factors such as language fluency and high versus low frequency words. The tool also integrated personal attributes, such as the size of a stroke and individual social support.
As part of her clinical research, Shreya Parchure provides speech therapy to patients at Penn, using study results to help guide and tailor individualized care.
“Incorporating language into the fold adds a new layer of considering human [and] brain complexity,” Parchure says, noting how the explainable AI tool was also able to predict speech performance word by word.
This granularity can help clinicians better uncover the underlying factors affecting a patient’s speech abilities and inform nuanced treatment and predicted recovery. What makes the model especially useful, Parchure says, is the ability to share the reasoning behind its recommendations.
“It’ll help tailor speech therapy for where exactly people are having trouble,” Parchure says. “We can really meet patients where they are in a more personalized manner.”
An innovative app for predicting recovery
Parchure and colleagues developed and launched an AI-powered app for use in clinical and research settings. The goal, she says, is to pilot the app in Hamilton’s aphasia clinic at Penn and determine its usefulness for personalized care.
Parchure’s work also has implications for improving knowledge of how aphasia presents in stroke survivors. “We have a better understanding of which parts of the brain, for example, are relevant to [post-stroke aphasia], so that might give new targets for brain stimulation or different therapies that we can individualize for a person’s needs,” she says.
A particularly innovative feature of this research, Parchure says, is that the AI model can simulate a “digital twin” for each patient, which functions as a predictive tool for language recovery. The simulated “twin” provides a comparative example of how a patient may respond to different treatments, which can elevate clinical trial efficiency by allowing researchers to compare projected versus actual recovery.
Translating lab findings into aphasia treatment
Growing up across two continents shaped Parchure’s understanding of language ability as central to quality of life, inspiring her varied interests. She credits bioengineering courses at Penn with informing her innovative approach to research: Classes on network neuroscience and brain-computer interfaces, she says, have not only complemented her medical school education, but also helped shape the machine learning aspects of the explainable AI study and improved her understanding of how to bring new findings from lab to bedside.
“The goal of my M.D.-Ph.D. training has been to translate advances in research in a way that will benefit patients,” says Parchure, who was awarded Best Poster in Translational Research at the 2025 PSOM Student Research Symposium.
Parchure anticipates that over the next decade, AI will play a pivotal role in personalizing speech therapy and help build a world in which every stroke survivor with aphasia can reconnect with the joy of language. One day, Parchure hopes to accelerate treatments for neurological conditions, treating patients in clinic and leading a bioengineering research lab of her own to benefit many more.
“I’ve been able to see a path where I can advance our understanding of complex diseases, and also use that new knowledge to really do good for society,” she says.
When the Schuylkill swallowed the city: Lessons from Hurricane Ida’s historic flood
New Penn research shows that Hurricane Ida wasn’t a once-in-a-century anomaly but a preview of how climate change, urbanization, and aging infrastructure are rewriting flood risk.
The Fed explained: What it does and why it matters
Former Philadelphia Fed President Patrick Harker and financial historian Peter Conti-Brown, both Wharton professors, unpack the central bank’s origins, its unusual structure, and the quiet ways it shapes the economy
How population changes are impacting primary education worldwide
Research from Penn sociologist Emily Hannum and colleagues reveals regional trends in whether school-age populations are increasing, plateauing, or decreasing—and shows how different countries are responding.
A bioengineered bean gum from the lab of Penn Dental’s Henry Daniell is found to reduce the levels of three microbes associated with head and neck squamous cell cancer to almost zero, without affecting the beneficial bacteria normally found in the mouth.
Fighting oral cancer with bioengineered chewing gum
Research led by Penn Dental’s Henry Daniell shows that antiviral and antibacterial chewing gums reduce the levels of three microbes linked to worse outcomes in oral cancers, paving the way for more effective and affordable therapies.
