ID #43 Exploring morphometric, lesion and texture-based analysis of MRI signatures in paediatric spinal cord tumours
Reeya Patel, Georgia Bright, Amanda Luck, Mikaela Doig, Hannah Rose Wardill, Jordan Hansford, Ryan O’Hare DoigAbstract
Background
The diagnosis of paediatric spinal cord tumours (SCTs) relies heavily on invasive biopsy, increasing the risk of morbidity and poor-quality survivorship. Magnetic resonance imaging (MRI), the gold standard for diagnosis and treatment planning, is rarely used in isolation due to lack of standardised analysis frameworks. We hypothesise that quantitative MRI approaches, including spinal cord morphometry and texture analysis, may capture subtle deformation and microstructural changes undetected by conventional pipelines, Our aim was to test scalable post-hoc pipelines which could support standardised analysis and biomarker-driven studies.
Methods
In this pilot study, retrospective data from diagnostic T2-weighted MRI were acquired from 8 paediatric patients aged 1-14 years with SCTs: astrocytoma (n = 3; female; cervicothoracic; low grade), ependymoma (n = 3; 2 male/1 female; thoracolumbar; low grade), diffuse midline glioma (n = 1; female; cervical; high grade), and Ewing sarcoma (n = 1; female; lumbar; high grade). Using Spinal Cord Toolbox, spinal cord segmentation and vertebral labelling enabled extraction of exploratory domains: morphometry, lesion metrics, and texture features derived from grey-level co-occurrence matrix. Analyses focused on descriptive statistics and exploratory correlation screening to identify preliminary patterns.
Results
Processing and semi-automatic segmentation succeeded across all scans. Tumour-affected segments showed increased cross-sectional area relative to unaffected tissue, with no consistent grey or white matter differences by tumour types or grade. Exploratory correlations suggested greater anterior-posterior cord angle in circumscribed tumours (correlation coefficient (r)=0.976, p = 0.005), ependymal lesions increased right-left diameter (r = 0.861, p = 0.028), and infiltrative segments reduced eccentricity (r = −0.760, p = 0.029). Axial damage ratio was lower in non-infiltrative tumours (r = −0.757, p = 0.030). Texture analysis revealed strong tumour presence with reduced contrast (r = −0.968, p = 1.0 × 10⁻6) and dissimilarity (r = −0.958, p = 3.7 × 10-6), while low-grade astrocytomas showed increased homogeneity (r = 0.848, p = 0.001) and energy (r = 0.734, p = 0.010).
Conclusion
Our preliminary findings demonstrate feasibility of quantitative MRI feature extraction in paediatric SCTs, revealing informative morphometric, lesion, and texture biomarkers.