Language outcomes after speech and language therapy in post-stroke aphasia are challenging to predict. This study examines behavioral language measures and resting state fMRI (rsfMRI) as predictors of treatment outcome. Fifty-seven patients with chronic aphasia were recruited and treated for one of three aphasia impairments: anomia, agrammatism, or dysgraphia. Treatment effect was measured by performance on a treatment-specific language measure, assessed before and after three months of language therapy. Each patient also underwent an additional 27 language assessments and a rsfMRI scan at baseline. Patient scans were decomposed into 20 components by group independent component analysis, and the fractional amplitude of low-frequency fluctuations (fALFF) was calculated for each component time series. Post-treatment performance was modelled with elastic net regression, using pre-treatment performance and either behavioral language measures or fALFF imaging predictors. Analysis showed strong performance for behavioral measures in anomia (R2 = 0.948, n = 28) and for fALFF predictors in agrammatism (R2 = 0.876, n = 11) and dysgraphia (R2 = 0.822, n = 18). Models of language outcomes after treatment trained using rsfMRI features may outperform models trained using behavioral language measures in some patient populations. This suggests that rsfMRI may have prognostic value for aphasia therapy outcomes.

Exploratory ascultation of brain signals has been prototyped in a project involving neurologists, real-time EEG and techniques for computer-based sound synthesis. In a manner similar to using a stethoscope, the listener can manipulate the location being listened to. Sounds which are heard are sonifications of electrode signals. We present a method for exploring sounds from arrays of sensors as sounds which are useful for distinguishing brain states. The approach maps brain wave signals to modulations characteristic of human voice. Computer-synthesized voices “sing” the dynamics of wakefulness, sleep, seizures, and other states. The goal of the project is to create a recognizable inventory of such vocal “performances” and allow the user to probe source locations in the sensor array in real time.