Psychology professor uses artificial intelligence and natural language processing to improve patient-caregiver relationship

A patient's mental health therapy notes might seem straightforward, but George Mason University professor and clinical psychologist Natasha Tonge believes those notes can be mined for cultural and social identifiers-subtle, sometime hidden, clues about a patients' life and experience-that can help inform a patient's care but are often missed in a busy clinical setting. 

Tonge is leading a research project, "Leveraging AI/ML to Improve Cultural Preparedness of Mental Health Professionals," that uses artificial intelligence and machine learning (AI/ML) to parse therapy notes to dig out those subtle clues. The project just received $64,000 from Morehouse School of Medicine on a subaward from the National Institutes of Health.  

"The goal of this project is to make it easier to identify a taxonomy of the different types of cultural and social identifiers in therapy records that may be relevant for patient care," said Tonge, who is a clinical health provider and the founder and director of George Mason's Trust and Interpersonal Disclosure Lab, located at Fuse at Mason Square.   

Examples of social identifiers include language that might give clues about a patient's culture and community background, gender identity, socioeconomic status, and more, Tonge explained.  

Tonge first began thinking about the concept of gaps in cultural identifiers and data collection in the mental health field while working at a psychiatric rehabilitation center in St. Louis, where she discovered that Black patients confided to her that they were having trouble connecting with the center's staff.  

To possibly bridge this treatment gap, Tonge turned to AI. After anonymizing patient treatment notes, "I test different strategies for identifying finding identifiers in the text in a process known as labeling, or annotation," she explained. 

The AI approach being utilized is a natural language processing pipeline. Once the manual labels are applied-for example, labeling a phrase like "I want to be a good mom" as an identifier for gender identity or labeling it as an indication of the social identity of being a parent-a SpaCyneural network pipeline (v 3.7.5) with a tok2vec encoder is trained on the labeled data.  

"I then compare the performance or impact of different labeling strategies on the algorithm's performance and how well it might learn new categories. I continue to experiment with labeling strategies and optimizing algorithm performance," she said. 

Tonge currently supervises George Mason graduate students in clinical psychology, working in the Center for Community and Mental Health(CCMH), George Mason's primary training clinic for doctoral candidates in clinical psychology. Tonge also works with a team of volunteers who pull and analyze CCMH data records of consenting patients.  

"The center is integrated into the community, and as a result we see a pretty wide range of people seeking therapy services," said Tonge.  

Rising junior Griffin Perry is working alongside Tonge in her lab as a summer project manager, overseeing the de-identification process.  

"We're putting the data we collect into the annotation software, Prodi.gy, to train the AI models," said Perry, who is double majoring in psychologyand community healthat George Mason.  

"I love [working with] Dr. Tonge because she bridges my two majors and areas of interest. These are the types of research areas and questions that lead to positive policy outcomes and changes, and it's actually what I'm doing for my thesis with her as well," said Perry, who transferred to George Mason's psychology program from Northern Virginia Community College through the ADVANCE Program.  

Perry, who is Tonge's advisee, will be doing an Undergraduate Research Scholars Programwith Tonge in the fall.  

"It's also important to train future clinicians on these emerging technologies so they can understand and apply them at a practical level, so they are able to deal with AI-related problems," said Perry, who will be working as a learning assistant for the College of Humanities and Social Sciencesthis fall.  

The newest Research at George Mason

Related Stories