Abstract A020: Decoding the solar elastosis microenvironment in melanoma with an interpretable multimodal fusion framework

Abstract The tumor microenvironment (TME) is a critical driver in the evolution of melanoma, which accounts for half of all skin cancer deaths despite its relative rarity. Solar elastosis, a pathological alteration of the dermal TME resulting from chronic UV exposure, is a key feature of this process and is closely associated with tumor mutation burden and patient outcomes. Surprisingly, melanoma patients with severe solar elastosis exhibit better survival rates than patients with mild solar elastosis. However, research into whether TME remodeling from sun exposure confers protective benefits for melanoma patients is hindered by difficulties in reliably characterizing solar elastosis, which suffers from high interobserver variability in histologic assessment and inconsistencies between clinical and histological findings. To address these challenges, we introduce HistoPath-Fusion, an interpretable solar elastosis predictor that integrates multiple-instance learning (MIL) with ensemble approaches. HistoPath-Fusion combines clinical data, histological images, and molecular profiles, hypothesizing that this multimodal fusion can improve accuracy and reliability. Our results show that while clinical and histology data independently yield reasonable predictions, their combination via ensemble methods offers only marginal gains. Notably, histology alone frequently outperforms multimodal models, indicating that the critical predictive features are largely captured within histological images. This suggests that histology represents the upper limit of predictive performance for solar elastosis, with clinicopathological variables contributing minimal additional value. Importantly, the HistoPath-Fusion interpretability framework supports pathologists by leveraging attention scores to highlight informative regions in whole slide images (WSIs). Pathologists showed strong concordance with HistoPath-Fusion in high-attention regions, achieving near 100% accuracy in identifying regions of severe and moderate solar elastosis, compared to only 8.4% concordance in low-attention areas. This significantly enhances diagnostic precision and reduces interobserver variability, demonstrating the utility of the HistoPath-Fusion framework for pathologists analyzing solar elastosis in cases of melanoma. Beyond diagnostic gains, HistoPath-Fusion reveals molecular insights, identifying downregulation in pathways related to skin development, lipid metabolism, and cytoskeletal integrity—suggesting impaired skin regeneration and barrier function from UV-induced damage. By accurately characterizing solar elastosis, the HistoPath-Fusion framework provides a powerful tool for dissecting the TME's role as a driver of tumor evolution and advancing disease diagnosis. Citation Format: Kushal Virupakshappa, Yue Hu, Kaibing Hu, Avinash Sahu. Decoding the solar elastosis microenvironment in melanoma with an interpretable multimodal fusion framework [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Cancer Evolution: The Dynamics of Progression and Persistence; 2025 Dec 4-6; Albuquerque, NM. Philadelphia (PA): AACR; Cancer Res 2025;85(23_Suppl):Abstract nr A020.