Pediatric Wilms tumor subtyping may skip biopsy: Dstar separates stromal from epithelial, mixed and regressive subtypes at slice level
Synopsis
Subtype differentiation in pediatric Wilms tumor now has a noninvasive imaging lead. Clinicians previously relied on the apparent diffusion coefficient, which showed no significant subtype differences across thirty children. A self-supervised implicit neural representation network estimated intravoxel incoherent motion parameters from ten b-value diffusion-weighted MRI, and a patient-specific 3D printed cutting guide aligned imaging with histopathology at slice level, showing significant Dstar differences between stromal and epithelial, mixed, and regressive subtypes.
Interpretation
Wilms tumor subtypes can be separated before surgery using MRI-derived parameters rather than invasive tissue sampling for histology. The conventionally used apparent diffusion coefficient showed no significant subtype differences in this cohort. Thirty lesions from thirty children were correlated between MRI and histopathology at slice level, with p=0.024 for stromal versus epithelial, p=0.031 for regressive, and p=0.043 for mixed tumors.
The discrimination comes from estimating diffusion and perfusion separately instead of using a single diffusion index. Intravoxel incoherent motion separates the diffusion coefficient D, incoherent blood flow or pseudodiffusion Dstar, and perfusion fraction f, whereas ADC yields one combined value. Parameters were estimated by a self-supervised implicit neural representation network from standardized 1.5T ten b-value diffusion-weighted imaging and matched to histopathology slice by slice using a patient-specific 3D printed cutting guide.
Perspective
The result rests on thirty children with Wilms tumor and thirty lesions in a retrospective single-center dataset, with imaging-histopathology correspondence at the macroscopic slice level, parameters from standardized 1.5T ten b-value diffusion-weighted imaging, and pairing via a patient-specific 3D printed cutting guide. The intended readers are researchers and clinical engineers working on pediatric solid tumor imaging subtyping and imaging-pathology correlation.
Subtype differences were significant only for Dstar, while ADC showed no significant differences; the findings come from a single-center retrospective design with a limited sample, the slice-level correspondence is macroscopic, and prospective validation or external cohort replication is not yet reported.