ADC cannot grade a renal cancer, and radiomics has not yet shown it can either

Design

systematic review with random-effects diagnostic meta-analysis

Population

20 studies of preoperative MRI in histopathologically proven renal cell carcinoma; 12 poolable

Primary outcome

diagnostic accuracy for WHO/ISUP high-grade disease

Effect

ADC sensitivity 0.84 (0.77–0.89), specificity 0.57 (0.51–0.63), AUC 0.71; radiomics/deep learning sensitivity 0.79, specificity 0.86, AUC 0.90

A systematic review searched four databases to August 2025 for quantitative MRI biomarkers predicting WHO/ISUP grade in histologically proven renal cell carcinoma. Twenty studies were included and twelve could be pooled — seven using apparent diffusion coefficient and five using MRI-inclusive radiomics or deep learning.

ADC-based grading gave pooled sensitivity 0.84 (95% CI 0.77 to 0.89) but specificity of only 0.57 (0.51 to 0.63), with a summary AUC of 0.71. Low-grade tumours did have higher ADC values than high-grade ones (mean difference 0.21 × 10⁻³ mm²/s, 95% CI 0.11 to 0.30). The radiomics and deep learning models performed better — sensitivity 0.79 (0.64 to 0.89), specificity 0.86 (0.74 to 0.93), AUC 0.90 — but across only five studies and with less consistency.

A specificity of 0.57 is the number that matters. It means nearly half of low-grade tumours are called high-grade, which is not a test you can put in front of a surgical decision. The authors ask for standardised multiparametric protocols, external validation and decision-impact studies before implementation, and that sequence is the right one — the radiomics AUC of 0.90 comes from five heterogeneous studies with no external validation behind it.

• Do not report an ADC value as evidence of tumour grade
• Use diffusion findings descriptively, and let biopsy or resection establish grade
• Treat radiomics grading models as research tools until externally validated
• Where a quantitative measure is quoted in a multidisciplinary meeting, say what its specificity is
• Standardise your renal MRI protocol — protocol variation is part of why these numbers do not converge

Why it matters

Quantitative values from these sequences are already appearing in reports as though they carried grading information.

Don't overread it

These are diagnostic accuracy estimates against histology; none of the studies tested whether using the biomarker changed a decision or an outcome.

The statistics, in plain English

Sensitivity 0.84 with specificity 0.57 means the test rarely misses a high-grade tumour but calls a great many low-grade ones high-grade — useful for ruling out, useless for ruling in. An AUC of 0.71 is modest discrimination. The radiomics AUC of 0.90 rests on five studies with wide confidence intervals and no external validation, which is exactly the pattern that shrinks when tested elsewhere.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.