In 228 chronic left-hemisphere stroke patients, normative tract disconnection predicted aphasia little better than a straight-fibre null; conduction aphasia followed posterior STG, not arcuate, damage

Question: does white-matter disconnection, estimated by projecting a lesion onto a normative tractogram, predict aphasia beyond lesion size and location, and does the fibre route matter? Approach: we registered the 228 manual lesion masks of the open Aphasia Recovery Cohort (OpenNeuro ds004884 v1.0.2) to MNI152NLin2009cAsym with dipy (brain-masked affine, enantiomorphic filling, SyN) and projected them onto the HCP-842 tractogram (144,678 streamlines, 78 bundles). As a control, we built a null tractogram with the same streamline end points but straight fibres. Outcomes were WAB-R aphasia quotient (n = 228), non-fluent versus fluent type (102 versus 94) and conduction versus anomic type (26 versus 61). We used two-stage ridge models with 10-fold cross-validation (100 repeats) and subject-bootstrap intervals, and a priori logistic tests. Findings: log lesion volume gave R2 = 0.366 for severity; parcel lesion load added 0.001, bundle disconnection 0.086 (0.033 to 0.139) and parcel-pair disconnection 0.067 (0.011 to 0.120), but the straight-fibre null also added 0.054 and 0.037; true minus null was 0.032 (-0.004 to 0.068) and 0.030 (-0.005 to 0.064). For fluency, bundles added 0.060 AUC to volume and the null 0.042; frontal aslant tract disconnection predicted non-fluency beyond frontal speech cortex damage (odds ratio 2.63 per SD, p = 0.0003) and beat its own straight-fibre null, but the frontopontine tract did equally well. For conduction aphasia, cortical location gave AUC 0.744 and pair disconnection 0.672, below its null (-0.059, -0.110 to -0.011); arcuate fasciculus disconnection did not predict conduction aphasia (odds ratio 1.24, p = 0.59) under three measures, while posterior STG damage did (3.23, p = 0.0001). Excluding 11 weaker registrations did not change these results. Limitations: normative not patient-specific tractography; template registration error; arcuate any-hit saturates in large lesions; ARC gives only AQ and WAB type, not sub-scores; 26 conduction cases, and arcuate and posterior cortex damage overlap. The registered masks, features, code and figures are included.

stroke · aphasia · lesion-symptom mapping · structural disconnection · normative connectome · tractography · null model · arcuate fasciculus · conduction aphasia · frontal aslant tract · verbal fluency · Aphasia Recovery Cohort · HCP-842 · cross-validation · MNI registration · lesion masks