An AI-powered Digital Therapy Assistant for Monitoring and Treating Cognitive Impairment

Investigators:
Jennifer Flexman, Moneta Health
Michael Busa, UMass Amherst

MassAITC Cohort: Year 3 (AD/ADRD)

Project Accomplishments: This project advances AI-based speech analysis to support the commercialization of Moneta Health’s virtual cognitive rehabilitation therapy platform, designed to help people living with dementia remain as independent as possible. In Aim 1, the team developed and validated models using ~100 hours of de‑identified therapy transcripts to score patient responses for synonym generation, antonym generation, and auditory comprehension tasks. Using Gemini LLMs and clinician‑labeled training data, the models achieved over 90% accuracy.

Aim 2 involved deploying these models in a real-world clinical study with 25 participants, confirming model reliability across demographic and cognitive subgroups, with no significant performance disparities detected. Sentiment and engagement markers were extracted from patient speech to support therapist decision-making, while survey response analysis provided additional insights into patient experience. Overall, the validated AI models demonstrated error rates below 15% and were deemed ready for deployment within Moneta’s platform to improve therapist efficiency and program scalability. The project also produced broader organizational outcomes, including Moneta’s $4.5M funding round, new healthcare partnerships, and multiple academic presentations and publications showcasing the technology’s impact and clinical potential. 

Initial Proposal Abstract: Cognitive rehabilitation therapy (CRT) is an established behavioral intervention led by skilled therapists to address challenges in daily life faced by individuals with mild cognitive impairment (MCI) and the early stages of dementia. Despite the evidence showing the ability of CRT to improve or preserve cognitive function, quality of life, and caregiver burden at the earlier stages of decline, 90% of seniors do not have access, particularly outside of skilled nursing settings and centralized medical centers. Moreover, CRT is typically limited to short time periods despite the progressive, long-term nature of underlying conditions such as Alzheimer’s disease.

Moneta Health is a brain health company reinventing the delivery of CRT to dramatically improve accessibility, particularly for communities with limited access to healthcare. The Moneta™ digital therapy assistant uses voice-based technology and artificial intelligence (AI) to deliver on-demand cognitive exercises over the telephone and analyze recorded speech for remote monitoring by speech therapists.

This project focuses on advancing the capabilities of the Moneta™ digital therapy assistant. Firstly, we will train and optimize the performance of AI algorithms to classify speech biomarkers relevant to monitoring patient response to therapy using existing speech data. Secondly, we will evaluate the usability and scalability of these AI algorithms through real-world testing with a cohort of older adults diagnosed with MCI or mild dementia (n=30), with the goal of achieving a clinically relevant predictive accuracy of >80%. The ability to accurately classify speech biomarkers enable real-time feedback for patients, and clinical decision support tools critical to commercialization.

Outcomes: