Introduction: Skin cancer diagnosis relies on histopathological evaluation of whole slide images (WSI), a task subject to inter-observer variability and time pressure. Cleo Skin (Primaa) is an AI-powered in vitro diagnostic software designed to support pathologists in lesion classification and mitosis detection. This study evaluated its impact on workflow efficiency under standardized reading conditions.

Methods/Design: SkinPerf2026 was a prospective multi-reader multi-case (MRMC) crossover study conducted at 11 French institutions. Sixteen pathologists reviewed 313 WSIs (240 for classification, 73 for mitosis detection) stained with HE/HES, both with and without Cleo Skin assistance, after a minimum 3-week washout period. Secondary endpoints included reading time, diagnostic confidence (10-point self-reported scale), inter-reader agreement (Fleiss' kappa), and immunohistochemistry (IHC) request rates. Statistical comparisons used Wilcoxon signed-rank tests.

Results: AI assistance significantly reduced reading time for both classification (median: 33 vs. 36 s, - 8%, p<0.001) and mitosis detection (median: 41 vs. 54 s, - 23%, p<0.001). Diagnostic confidence increased with AI for classification (mean: 8.58 vs. 8.23, p<0.001) and mitosis detection (mean: 8.67 vs. 8.27, p<0.001). Inter-reader agreement improved from substantial to almost perfect for classification (Fleiss' κ: 0.86 vs. 0.81, p<0.01), and from moderate to substantial for mitosis detection (κ: 0.66 vs. 0.56). IHC requests decreased with AI assistance (mean 182 vs. 217 slides per pathologist).

Conclusion: Cleo Skin was associated with meaningful improvements in workflow efficiency, reader consistency, and diagnostic confidence across both classification and mitosis detection tasks. These findings support the integration of AI decision-support tools into routine dermatopathology workflows to reduce pathologist burden without compromising diagnostic performance.

Learning Objectives:

1. Describe the impact of AI assistance on reading time and diagnostic confidence in digital dermatopathology
2. Recognize how AI tools reduce inter-reader variability in skin lesion classification and mitosis detection
3. Discuss the potential of AI-assisted workflows to streamline IHC decision-making in routine pathology practice