DIASURV · EJC SKIN CANCER

Comparison of an AI-Based Whole-Slide Image Prognostic Assay with Gene Expression Profiling in Early-Stage Melanoma

C. Bossard, Y. Salhi, A. Khammari, M. Brousseau, Y. Le Corre, S. Salhi, G. Quéreux, J. Chetritt

2026

Abstract

Background

Gene expression profiling (GEP) assays have been introduced to refine prognostic stratification in early-stage melanoma, but their clinical utility remains debated. DiaSurv Melanoma is an AI-based prognostic assay applied to routine hematoxylin–eosin (H&E) whole-slide images (WSI) to stratify risk without additional molecular testing. We evaluated DiaSurv Melanoma (DiaDeep, France) against a commercially available GEP assay in a retrospective early-stage melanoma cohort.

Methods

A retrospective cohort of 50 early-stage melanoma patients, predominantly stage IA, was analyzed. Each case was independently classified as low-risk or high-risk by DiaSurv Melanoma and by the GEP assay. Concordance between the two approaches was assessed. Overall survival (OS) and relapse-free survival (RFS) were evaluated, and predictive performance was summarized using positive predictive value (PPV) and negative predictive value (NPV). Kaplan–Meier survival analyses with log-rank testing were performed for both assays.

Results

Overall concordance between DiaSurv Melanoma and the GEP assay was 74%. For OS, most deaths occurred in patients classified as high-risk by DiaSurv Melanoma (5/7; 72%), compared with those classified as high-risk by GEP (2/7; 29%). DiaSurv achieved PPV/NPV of 26%/94% for OS, versus 20%/88% for GEP. For RFS, DiaSurv Melanoma assigned all relapses (4/4) to the high-risk group, whereas the GEP assay classified only half of these cases as high-risk. Corresponding PPV/NPV were 21%/100% for DiaSurv and 20%/95% for GEP. Kaplan–Meier analyses demonstrated a significant separation of risk groups for RFS with DiaSurv Melanoma (log-rank p < 0.05), while no significant difference was observed with the GEP assay.

Conclusion

In this retrospective cohort of early-stage melanoma, DiaSurv assay showed prognostic performance comparable to, and in some aspects superior to, gene expression profiling. DiaSurv Melanoma more accurately identified patients at risk of relapse and death, while requiring no additional tissue, molecular testing or specialized workflows. These results support further validation of AI-based digital pathology as a practical alternative or complement to molecular prognostic assays in melanoma care.

Read the full paper

Try it on your cases ↗︎