Document resource
Introduction No proven subgroup of basal ganglia intracerebral hemorrhage (bgICH) patients benefits from minimally invasive surgery (MIS). Corticospinal tract (CST) integrity may guide surgical selection, but requires MRI with diffusion tensor imaging, which is impractical acutely. We evaluated whether automatically generated CT-derived CST biomarkers could identify bgICH patients uniquely benefiting from surgery.Materials and Methods From a previously published retrospective single-center propensity-matched cohort of moderate-sized (10-50 mL) putaminal ICHs, the last preoperative CT (surgical) or 24-hour stability CT (medical) was analyzed. A 3D nnU-Net trained on 307 DTI-CT pairs generated bilateral CST masks from CT alone (Dice 0.718). ICH masks were generated using a validated segmentation model. CSTs were split into ipsi- and contralesional hemispheres, and 93 radiomic texture features were extracted from the ipsilesional CST restricted to hemorrhage-bearing axial slices, each normalized to the contralesional CST as an internal control. Each feature quantified structural asymmetry between tracts, where greater asymmetry reflects hemorrhage-related tract disruption. Treatment-by-feature interactions for mRS 0-2 were tested via logistic regression with likelihood ratio tests and Benjamini-Hochberg false discovery rate (FDR) multiple comparisons correction at q < 0.20.Results Sixty-two patients (31 surgical, 31 medical, 56% male, median age 57 years) were included. Cohorts were balanced on GCS (median 12 vs 12, p = 0.854), NIHSS (18 vs 17, p = 0.724), and IVH (29% vs 29%, p = 1.0); surgical patients had larger ICH volumes (30.5 vs 22.3 mL, p = 0.040). Median time from last-known-well to evacuation was 12 hours (IQR 7.2-18.8) with a postoperative residual of 2.4 mL (IQR 0.7-6.7). Overall, mRS 0-2 rates were similar (surgical 10% [3/31] vs medical 13% [4/31], p = 1.0), while mortality trended lower with surgery (3% [1/31] vs 23% [7/31], p = 0.053). Sample CST and bgICH segmentation is shown in Video 1. ( https://drive.google.com/file/d/1K5fMZ_fFSiUsJSX0VGLYHa_bGGsMDlEU/view?usp=sharing). Two features reflecting CT attenuation heterogeneity within the ipsilesional CST demonstrated significant treatment interactions for mRS 0-2 surviving FDR correction (LRT p = 0.0019, q = 0.10 and LRT p = 0.0022, q = 0.10). Surgery was favored when the ipsilesional tract retained heterogeneous signal, while no benefit was observed when the tract showed homogeneous attenuation (figure 1).Conclusion CT-derived CST features can identify bgICH patients more likely to respond with MIS evacuation. These findings warrant validation in multicenter prospective studies and may enrich future bgICH surgical trials.Disclosures A. Kashkoush: None. S. Li: None. D. LIlly: None. E. Plow: None. S. Raymond: 4; C; VonVascular, Magnendo, Kannact.Abstract E-148 Figure 1(A) Representative axial CT with ipsilesional CST (red), contralesional CST (blue), and ICH (yellow) overlay. (B) Good functional outcome (mRS 0-2) rates stratified by median CT-derived CST structural integrity and treatment group. Surgery was favored when the ipsilesional CST retained heterogeneous attenuation (high integrity) (interaction LRT p = 0.0019, FDR q = 0.10)