Mapping thrombus habitat: Non-contrast MRI radiomics and pixel-tile histomics approach to track venous thrombosis evolution in mice.

BACKGROUND

Determining the age and composition of deep vein thrombosis (DVT) is essential for optimizing thrombus removal strategies, yet clinical decision-making remains restrained by the subjectivity of symptom-based assessment. We applied non-contrast T1/T2 weighted MRI radiomics with habitat clustering to quantify thrombus evolution, validating MRI habitats against histopathology with translational clinical potential.

METHODS

Venous thrombosis was induced in BALB/c mice using a refined inferior vena cava ligation model to ensure consistent thrombus development. Longitudinal and cross-sectional cohorts were imaged at 7 T MRI with T1- and T2-weighted sequences on post-ligation days 2, 7, and 14. Thrombus regions were delineated using a hybrid approach and voxel-level radiomic features were extracted. Unsupervised K-Means clustering defined MRI-based habitats. Imaging signatures were correlated with an artificial neural network pixel classifier and tile-based histomics clustering to quantify red blood cells (RBC), fibrin, and collagen.

RESULTS

Radiomic analysis showed a decline in Energy and TotalEnergy and a similar trend in GrayLevelNonUniformity, reflecting a transition from acute to more uniform chronic thrombi. A two habitats strategy could differentiate acute RBC-rich regions from fibrotic components. Histology confirmed RBC-rich areas fell from 88.22% to 6.52% over 14 days, while collagen-rich areas rose from 0.16% to 79.67%. MRI habitat proportions correlated strongly with histological tile clusters (Spearman r = 0.74, p = 0.0005).

CONCLUSION

Multiparametric MRI radiomics combined with habitat analysis captures thrombus composition and temporal evolution, with unsupervised histomics providing biological validation. These findings support the translational potential of MRI habitat analysis as a non-invasive biomarker of thrombus heterogeneity.