Comparing human-whole slide image (WSI) and AI-WSI assessment of tumor immune microenvironment in pleural mesothelioma: Insights from the ANEMONE Project.

Abstract

e20096 Background: Tumor immune microenvironment (TIM) is critical in cancer progression and therapeutic response. In pleural mesothelioma (PM), the limited efficacy of immune checkpoint inhibitors underscores the need for precise and reproducible TIM characterization to identify predictive biomarkers and improve patient stratification. However, no universally accepted method exists for a standardized evaluation of the TIM. This study aimed to evaluate the concordance between human- whole slide image (WSI) and AI-WSI assessment for the quantification of TIM in clinical PM samples, focusing on the reproducibility and reliability of these methods. Methods: This preliminary report presents findings from the ANEMONE project a multicenter prospective and retrospective study funded by the European Commission under the TRANSCAN initiative, involving five specialized European centers. Immunohistochemical staining was performed for key TIM markers (CD4, CD8, CD3, CD68, CD163, CD20, CD57). Human-WSI semi-quantitative evaluation was performed independently by two pathologists and Cohen’s Kappa (κ) agreement was calculated. AI-WSI scoring was carried out by a dedicated software (Visiopharm). Further comparison of quantitative results was conducted using Absolute Standardized Mean Differences (ASMD), with a threshold of ≤0.1 indicating high concordance. Results: A total of 153 patients were included in the study; 25 were recruited prospectively. The interobserver agreement by two pathologists was fair to moderate (Cohen’s Kappa range 0.21-0.57) across all markers. On ASMD analysis CD8 demonstrated the highest concordance between human-WSI and AI-WSI, (ASMD = 0.0735), followed by CD4, both of which below the threshold of 0.1 (Table 1). For all other markers, AI-WSI systematically underestimated marker expression compared to human-WSI scoring, a trend consistently observed across all markers. Conclusions: The variability in pathologists' assessments highlights the complexity of TIM evaluation, while AI systematic underestimation suggests limitations in handling staining and tissue heterogeneity. Standardizing pre-analytical phases could improve both methods. The concordance for CD8 and CD4 indicates that, with further refinement, AI may serve as a useful complementary tool for TIM analysis. ASMD values for immunohistochemical markers in TIM analysis. Immunohistochemistry marker ASMD CD20 0.284 CD3 0.237 CD68 0.209 CD163 0.139 CD57 0.130 CD4 0.103 CD8 0.654