A Novel Vit-Based Multi-Scaled and Rotation-Invariance Approach for Precise Differentiation Between Meningioma and Solitary Fibrous Tumor

Meningioma (MEN) and Solitary Fibrous Tumor (SFT) pose challenges in diagnosis due to their similar histological characteristics. Common diagnostic methods like immunohistochemistry (IHC) have drawbacks in terms of time, cost, and accessibility. This paper introduces a new approach using Vision Transformer (ViT) on whole-slide images (WSI) for MEN and SFT classification, addressing these challenges. The proposed model employs a rotational and scale invariance approach with a ViT architecture, capturing global and local features at various optical scales in WSIs. To comprehensively an-alyze, pre-trained ViT models (ViT B/16, B/32, L/16, L/32) were applied to different WSI magnification levels. Diagnosis was determined by combining WSI scales using a weighted majority voting scheme. Evaluation on a 92-patient cohort (46 MEN, 46 SFT) showed ViT L/16 model’s robust performance with 90.45% accuracy, 93.03% sensitivity, 87.94% specificity, 88.21% precision, 90.48% balanced accuracy, 90.62% ROC, and 90.56% F1-score. These results affirm the effectiveness of the proposed model, promising innovative tools in histopathology diagnostics.