Identifying neuropathologic disease in primary progressive aphasia using narrative speech

We present an application of artificial intelligence to narrative speech with the primary objective of predicting neuropathologic disease underlying primary progressive aphasia (PPA).

Using natural language processing toolkits, features were extracted from transcribed narratives of the Cinderella story. Machine learning ensemble models classified participants as either normal controls (NC) or as individuals with PPA and a subsequent autopsy-confirmed neuropathologic diagnosis of either Alzheimer's disease (AD) or 4-repeat tauopathy under the umbrella of frontotemporal lobar degeneration (FTLD-4Rtau).

All models successfully distinguished transcribed narratives of AD from those with FTLD-4Rtau, as well as the narratives of NC from those with PPA. Feature permutation revealed diverging patterns of contribution to classification depending upon language domain and disease pathology.

The usage of artificial intelligence in the context of naturalistic language tasks may ultimately serve as a complementary aid in differential diagnosis of PPA disease pathologies and in uncovering avenues for disease-specific interventions.