Abstract 2007: Prognostic value of immune infiltrating lymphocytes in primary cutaneous melanoma: Insights from the D-ESMEL study

Abstract Introduction: Most stage I/II cutaneous melanoma patients have favorable outcomes, while a subset experiences distant metastasis, highlighting limitations in current staging systems, which rely primarily on Breslow thickness and ulceration status. In recent years, the involvement of tumor-infiltrating lymphocytes (TILs) in the tumor microenvironment (TME) has gained attention for their prognostic significance in melanoma, while their role and reproducibility remain debated. Here, we developed and validated an artificial intelligence (AI) based model to segment histological features in whole-slide images (WSIs) of cutaneous melanoma. We quantitatively assessed multiple features, including TME lymphocyte involvement, and evaluated their prognostic value in metastasis prediction. Methods: The D-ESMEL study, a retrospective case-control study, included a discovery set (n=440, comprising 220 case-control sets matched based on key prognostic factors) with 46% stage I and 54% stage II melanomas) and a validation cohort (population-based cohort of 5,551 stage I/II patients, nested case-control set: n=308, consisting of 154 patients with distant metastasis and 154 controls). U-Net and DeepLab3+ neural networks, integrated into the Oncotopix® platform, were trained and validated using independent datasets to automate the segmentation of histological structures, including tumor region, TME, ulceration, and mitotic activity. The model was then applied to the D-ESMEL samples. Multivariable conditional logistic regression was performed to identify features associated with metastasis risk, and model performance was assessed in the validation cohort using the concordance index (C-index) and likelihood ratio test (LTR). Results: Our model segmented histological features with an accuracy surpassing 92%. In the discovery set, immune cell proportions were significantly lower in metastasis cases than in matched controls (median difference: -10.00%; 95% CI: -15.00% to -5.00%, p < 0.001). Higher TIL proportions were associated with reduced metastasis risk (odds ratio per 10% increase: 0.98; 95% CI: 0.96-0.99). Other features, including tumor area, density, and mitotic rate, showed no significant prognostic value. In the validation cohort, TIL proportions remained predictive when combined with Breslow thickness, ulceration, age, and sex (LRT: Chi-squared = 5.04, df = 1, p = 0.02) Conclusions: Our AI-driven model enables accurate segmentation and analysis of melanoma features in WSIs. TILs were the only feature reliably associated with metastasis risk, validated in an independent cohort, emphasizing their potential as a biomarker for risk stratification. while other tumor characteristics, such as tumor area, density, and mitotic activity—despite their known histopathological relevance—did not show predictive associations with metastasis risk in the D-ESMEL cohort. Citation Format: Thamila Kerkour, Loes Hollestein, Alex Nigg, Catherine Zhou, Tamar Nijsten, Yunlei Li, Antien Mooyaart. Prognostic value of immune infiltrating lymphocytes in primary cutaneous melanoma: Insights from the D-ESMEL study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2007.