Performance Models Do Not Improve Lesion Models in Predicting Post-Stroke Aphasia

Performance models based on structural and functional connectivity are not superior to those based on lesion location alone in predicting neurocognitive network-level disruption in patients with chronic post-stroke aphasia. These are the findings of a study published in the journal Brain.

Researchers in the United Kingdom and the United States conducted an in vivo study to test the functional and structural connectivity, enrolling 70 eligible patients with chronic post-stroke aphasia. By study conclusion, the researchers assessed 68 patients with structural magnetic resonance imaging (MRI) scans and 39 patients with functional MRI (fMRI) scans to determine whether current prediction models of performance based solely on lesion data might be improved using multiple modalities evaluating network-level connectivity.

The researchers devised 4 possible prediction models based on key language-cognitive factors: executive function, fluency, phonology, and semantics.

Patients underwent a battery of neuropsychological and cognitive tests to assess each patient’s cognitive function, fluency, and ability to comprehend sentences and process phonological input/output and semantics.

The researchers gathered neuroimaging data, including resting state fMRI, high resolution structural T1-weighted imaging, and diffusion weighted imaging to assess whether these images could predict behavioral factors and aphasia.

When analyzing structural connectivity, T1 lesion maps predicted 3 out of the 4 models, including phonology (P =.001), semantics (P =.004), and fluency (P =.001), but not executive function. Lesion clusters located at the superior temporal gyrus extending to the supramarginal gyrus corresponded to phonology. Lesions of the medial and anterior temporal lobe and inferior longitudinal fasciculus white matter correlated with semantics. Lesions of the precentral and postcentral gyri and white matter of the corticospinal tract and frontal aslant tract corresponded to fluency. The researchers were unable to find stable clusters correlating with executive function.

When analyzing functional connectivity, they discovered that regions of interest based on the structural T1 findings correlated with functional brain activity for phonology, semantics, and fluency as seen on fMRI scans.

Both structural and functional connectivity accurately predicted behavior in patients with chronic post-stroke aphasia. However, when the researchers compared these new predictive models to current lesion-only based models informed by T1 imaging, the new prediction models did not perform any better than the lesion-based models.

The researchers noted, “Overall, our findings suggest that accounting for lesion location is fundamental in understanding whether/how connectivity measures influence prediction models.”

“Together, the results indicate that the observed network-level disruption was predicted by the lesion alone and does not significantly improve model performance in predicting the profile of language impairment,” they concluded.

Study limitations included the practical issues that restrict analysis of prognosis, such as lack of range of stroke severity in study samples, differences in testing times and functional imaging times for patients, insufficient behavioral testing data, and generally smaller sample size in longitudinal studies.