Abstract
Background
Immune checkpoint inhibitors (ICIs) have transformed cutaneous melanoma therapy, yet 50% of patients show primary or acquired resistance, and current biomarkers (PD-L1 and tumor mutational burden) lack precision. High-frequency ultrasound (HFUS), contrast-enhanced ultrasound (CEUS), and shear-wave elastography (SWE) are inexpensive and repeatable, but their relationship to the immune transcriptome and ICI outcomes is unknown.
Methods
In this single-center retrospective study, 218 patients with cutaneous melanoma who received first-line anti-PD-1-based ICIs and pretreatment ultrasound were divided into training (n = 153) and internal validation (n = 65) sets. HFUS/CEUS/SWE radiomic features were filtered for reproducibility and selected by LASSO to construct an ultrasound radiomic score (US-score). Across three public transcriptomic cohorts (TCGA-SKCM, GSE91061, and GSE78220) and a melanoma single-cell dataset (GSE115978), we combined differential expression, WGCNA, immune deconvolution, and LASSO-Cox to derive an IRGS. Four core genes (CXCL10, CD8A, IFNG, and CXCL9) were validated
in vitro
by RT-PCR in three ATCC melanoma cell lines after IFN-γ stimulation.
Results
The objective response rate was 42.7% (median follow-up 28.4 months). The US-score independently predicted response (adjusted OR: 2.6; 95% CI: 1.7–3.9; validation AUC: 0.78). The IRGS stratified public cohorts into immune-hot/cold groups with distinct outcomes (TCGA-SKCM overall-survival HR: 2.14; 95% CI: 1.56–2.93; response AUC: 0.79). The US-score correlated with the IRGS (r = 0.58) and with CD8
+
T-cell density on multiplex immunohistochemistry (r = 0.62). The combined nomogram outperformed any single biomarker (validation AUC: 0.88 vs. 0.78 [US], 0.76 [IRGS], and 0.68 [clinical]) with good calibration and net benefit and separated progression-free survival (median: 18.6 vs. 6.3 months; HR: 3.12, 95% CI: 2.08–4.68).
In vitro
, all four genes were significantly induced by IFN-γ (CXCL10 up to 8.3-fold; all
p
Conclusion
Ultrasound radiomics encodes information about the melanoma tumor-immune microenvironment; combined with an immune-related gene signature, it provides a non-invasive, biologically anchored tool for stratifying ICI response and resistance.