INS NYC 2024 Program

Poster	Poster Session 04 Program Schedule 02/15/2024 12:00 pm - 01:15 pm Room: Shubert Complex (Posters 1-60) Poster Session 04: Neuroimaging \| Neurostimulation/Neuromodulation \| Teleneuropsychology/Technology Final Abstract #27 Examining Speech Characteristics in Verbal Fluency using Large Language Models: Insights from an EMA study in a clinical population Shifali Singh, McLean Hospital \| Harvard Medical School, Belmont, United States Katelin Curtis, McLean Hospital \| Harvard Medical School, Belmont, United States Carla Agurto, Digital Health IBM Research, Yorktown Heights, United States Susan Waisbren, Boston Children’s Hospital \| Harvard Medical School, Boston, United States Guillermo Cecchi, Digital Health IBM Research, Yorktown Heights, United States Laura Germine, McLean Hospital \| Harvard Medical School, Belmont, United States Raquel Norel, Digital Health IBM Research, Yorktown Heights, United States Category: Teleneuropsychology/ Technology Keyword 1: phenylketonuria Keyword 2: speech Keyword 3: technology Objective: In this study, we examined verbal fluency through speech ecological momentary assessment (EMA) in a patient population diagnosed with Phenylketonuria (PKU). PKU is a rare, chronic condition caused by the buildup of Phenylalanine (Phe) in the body, resulting in patients experiencing ongoing problems with executive function, verbal fluency, and processing speed. In this patient population, aspects of speech characteristics are related to fluctuations in blood levels. Limited research exists in understanding the impact of fluctuations in Phe levels on speech, therefore, our objective was to establish potential connections between speech attributes and blood levels in patients with PKU. By identifying the nuances in speech fluctuations, we aim to better understand how specific fluctuations are associated with speech characteristics. Participants and Methods: We recruited twenty participants through the National PKU Alliance (NPKUA) to examine verbal fluency and fluctuations through speech EMA and finger prick tests. Eligible participants were invited to participate in an EMA framework that included completing a series of brief speech tests (6 EMAs) across a 30-day period. PerkinElmer provided participants with kits to self-collect a few drops of blood as well as analyze the blood. Blood samples were taken on the same days as the speech assessments to track Phe, Tyrosine (Tyr), and Phe:Tyr (ratio) blood levels. Speech EMA scores were compared with the blood samples to look at speech characteristics in verbal fluency using Large Language Models (LLMs). Results: We conducted three distinct analyses of speech characteristics, specifically focusing on verbal fluency. Analyses included 1) examining the association between spoken time and Phe:Tyr blood levels; 2) assessing the mean dispersion of words and Phe:Tyr ratios blood levels; and 3) evaluating semantic similarity using LLMs in relation to Phe:Tyr ratio blood levels. Overall, patients’ speech patterns during verbal fluency tests, including spoken time and dispersion of words, showed significant correlations (Spearman correlation; r=0.24, r=0.53, r=0.30 respectively, all with p<0.05) with Phe:Tyr ratio blood levels. Using LLMs, we observed that individuals with better Phe:Tyr ratio levels demonstrated proficiency in producing semantically related words. Conclusions: This study represents the first of its kind to utilize LLMs in the context of speech EMA among patients affected by PKU. The findings of this study provide insights into the effect of speech on a PKU population. With these implications, we can draw diagnostic conclusions that by looking closely at verbal fluency, there is an impact of fluctuations on speech. The outcomes support the potential for further exploration into utilizing speech patterns and optimal verbal efficiency as a method of ascertaining Phe:Tyr ratio blood levels in patients with PKU. The data acquired can be used to improve the diagnosis, monitoring, accessibility, and treatment of patients with PKU.

Poster

02/15/2024
12:00 pm - 01:15 pm
Room: Shubert Complex (Posters 1-60)

Poster Session 04: Neuroimaging | Neurostimulation/Neuromodulation | Teleneuropsychology/Technology

Final Abstract #27

Examining Speech Characteristics in Verbal Fluency using Large Language Models: Insights from an EMA study in a clinical population

Shifali Singh, McLean Hospital | Harvard Medical School, Belmont, United States
Katelin Curtis, McLean Hospital | Harvard Medical School, Belmont, United States
Carla Agurto, Digital Health IBM Research, Yorktown Heights, United States
Susan Waisbren, Boston Children’s Hospital | Harvard Medical School, Boston, United States
Guillermo Cecchi, Digital Health IBM Research, Yorktown Heights, United States
Laura Germine, McLean Hospital | Harvard Medical School, Belmont, United States
Raquel Norel, Digital Health IBM Research, Yorktown Heights, United States

Category: Teleneuropsychology/ Technology

Keyword 1: phenylketonuria
Keyword 2: speech
Keyword 3: technology

Objective:

In this study, we examined verbal fluency through speech ecological momentary assessment (EMA) in a patient population diagnosed with Phenylketonuria (PKU). PKU is a rare, chronic condition caused by the buildup of Phenylalanine (Phe) in the body, resulting in patients experiencing ongoing problems with executive function, verbal fluency, and processing speed. In this patient population, aspects of speech characteristics are related to fluctuations in blood levels. Limited research exists in understanding the impact of fluctuations in Phe levels on speech, therefore, our objective was to establish potential connections between speech attributes and blood levels in patients with PKU. By identifying the nuances in speech fluctuations, we aim to better understand how specific fluctuations are associated with speech characteristics.

Participants and Methods:

We recruited twenty participants through the National PKU Alliance (NPKUA) to examine verbal fluency and fluctuations through speech EMA and finger prick tests. Eligible participants were invited to participate in an EMA framework that included completing a series of brief speech tests (6 EMAs) across a 30-day period. PerkinElmer provided participants with kits to self-collect a few drops of blood as well as analyze the blood. Blood samples were taken on the same days as the speech assessments to track Phe, Tyrosine (Tyr), and Phe:Tyr (ratio) blood levels. Speech EMA scores were compared with the blood samples to look at speech characteristics in verbal fluency using Large Language Models (LLMs).

Results:

We conducted three distinct analyses of speech characteristics, specifically focusing on verbal fluency. Analyses included 1) examining the association between spoken time and Phe:Tyr blood levels; 2) assessing the mean dispersion of words and Phe:Tyr ratios blood levels; and 3) evaluating semantic similarity using LLMs in relation to Phe:Tyr ratio blood levels. Overall, patients’ speech patterns during verbal fluency tests, including spoken time and dispersion of words, showed significant correlations (Spearman correlation; r=0.24, r=0.53, r=0.30 respectively, all with p<0.05) with Phe:Tyr ratio blood levels. Using LLMs, we observed that individuals with better Phe:Tyr ratio levels demonstrated proficiency in producing semantically related words.

Conclusions:

This study represents the first of its kind to utilize LLMs in the context of speech EMA among patients affected by PKU. The findings of this study provide insights into the effect of speech on a PKU population. With these implications, we can draw diagnostic conclusions that by looking closely at verbal fluency, there is an impact of fluctuations on speech. The outcomes support the potential for further exploration into utilizing speech patterns and optimal verbal efficiency as a method of ascertaining Phe:Tyr ratio blood levels in patients with PKU. The data acquired can be used to improve the diagnosis, monitoring, accessibility, and treatment of patients with PKU.