Continuous-time random walk MRI and pathological heterogeneity in intracranial tumours: a neuronavigation-guided study

Abstract

Brain tumours are spatially heterogeneous, whereas MRI and histopathology interrogate tissue at different spatial scales. We developed a neuronavigation-guided radiology-pathology framework to test whether continuous-time random walk (CTRW) diffusion metrics measured at recorded sampling sites are associated with pathology-based structural heterogeneity and selected immune-cell proportions. We studied 98 surgically treated adults with high-grade glioma (HGG; n = 34), brain metastasis (BM; n = 34), or meningioma (MENI; n = 30). Each participant contributed one neuronavigation-recorded specimen of approximately 2 mm from the superficial margin within the solid contrast-enhancing tumour component, with a coordinate-centred 2-mm imaging ROI used for radiology-pathology matching. CTRW spatial heterogeneity (β), temporal heterogeneity (α), and anomalous diffusion coefficient (Dm) were derived from high-b-value DWI. Matched specimens underwent haematoxylin & eosin staining and immunofluorescence analysis. Pathologist scoring, pathomics, and immune-cell measurements were analysed at the patient level. Five regression families were compared using repeated tumour-type-stratified nested cross-validation. In malignant tumours, β decreased from contralesional brain to peritumoral brain oedema and contrast-enhancing tumour and was inversely associated with the manual H&E heterogeneity score (Spearman ρ = -0.948; FDR-adjusted q < 0.001). Random-forest regression had the highest internal cross-validation performance. In the 50-outer-fold sensitivity analysis, mean R² ± SD values were 0.984 ± 0.010 for neutrophils, 0.976 ± 0.013 for macrophages, and 0.943 ± 0.033 for B cells. Tumour-type-only models had lower mean R² values of 0.352, 0.295, and 0.320, whereas CTRW models retained mean R² values of 0.717, 0.767, and 0.671 after training-fold residualization for tumour type. At neuronavigation-recorded sampling sites, CTRW metrics were associated with pathology-defined structural heterogeneity and enabled internally validated estimation of selected immune-cell proportions. These findings support CTRW as a complementary imaging descriptor, although its incremental value beyond conventional diffusion metrics remains to be established. Importantly, the observed associations reflect radiology–pathology correspondence at the sampling-site scale rather than direct cellular-resolution imaging or validated clinical prediction. Nevertheless, the biologically anchored relationships identified here provide a rationale for the prospective evaluation of CTRW-derived imaging biomarkers for imaging-informed tumour characterization and stratification.