Automated Speech Analysis in FTD Spectrum Disorders
Abstract
Although half of patients with frontotemporal degeneration (FTD) have changes in social behavior and personality, the remaining half of FTD patients have a speech and language disorder known as primary progressive aphasia (PPA). This interferes with communication, perhaps the greatest challenge for FTD patients and their caregivers, reducing safety and increasing caregiver burden. There are different forms of PPA, and these are reasonably accurate at predicting the underlying pathology that causes PPA. Therefore, in an era of disease-modifying treatment trials, it is essential to have valid, objective, sensitive, reliable, and reproducible measures of speech and language that can facilitate diagnosis and monitor severity. Indeed, with the ability to assess speech and language longitudinally, these measures can be used in treatment trials to determine if the treatment is effective. We propose to adapt modern technology of speech recognition and natural language processing to develop novel, automated, and quantifiable biomarkers of FTD spectrum disorders using natural connected speech samples. We obtain these by asking patients to describe a picture so that we know the intended content of their speech, and we digitally record the speech. These natural biomarkers include acoustic elements such as intonation, rhythm, pauses, and stresses that are very difficult to judge objectively and reliably. These supplement the word content with linguistic meaning (e.g., the difference in stress between the noun “récord” versus the verb “recórd”), syntactic structure (e.g., the difference between the statement “you are tired!” versus the question “you are tired?”), and emotional overlay. Our automated algorithms also quantify word meaning and part of speech. Together with acoustic quantification, these assessments of words are highly sensitive and can detect the earliest signs of FTD in an inexpensive and non-invasive way, can diagnose the specific type of PPA and its associated pathology in a reasonably reliable manner, and can quantify objectively how speech and language are worsening. While these attributes of speech and language can be judged manually, this is extremely time-consuming and often unreliable since different judges have different opinions about the same speech patterns. Moreover, these natural speech and language markers are easy to collect repeatedly, non-invasively, and inexpensively. We will validate our novel algorithms for analyzing acoustic and lexical markers by relating them to biological indices through brain neuroimaging such as MRI (magnetic resonance imaging), biofluid markers such as cerebrospinal fluid and blood, and genotyping. Our cohort consists of a very large series of well-characterized patients with various forms of FTD disorders, including patients with aphasia (a language impairment), social-behavioral syndrome (behavioral variant FTD) and accompanying motor diseases (some with Parkinsonism such as progressive supranuclear palsy, others with amyotrophic lateral sclerosis). These disorders are included in the Fiscal Year 2019 Peer Reviewed Medical Research Program Topic Areas, and military combat personnel appear to be at a higher risk to develop them. Based on our previous publications and preliminary analyses, we hypothesize that specific measurable markers in the speech patterns of patients with FTD spectrum disorders can be collected and analyzed in an automated, objective, reliable, and reproducible manner, that these will differentiate between the different clinical presentations of PPA and FTD, that these specific patterns will be related to disease in specific brain regions in MRI studies, that abnormal speech patterns will have a predictive value for the underlying pathology, and that these speech patterns will have prognostic value for the longitudinal course of disease progression.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 10, 2021
- Source ID
- W81XWH2010531
Entities
People
- Murray Grossman
Organizations
- United States Army
- University of Pennsylvania