Accurate prostate cancer detection from Whole-Slide Images (WSIs) is critical for improving diagnostics. This study benchmarks 10 foundation models as feature extractors for graph-based classification using Graph Attention Networks (GATs). We compare them to CNN-based extractors, including ResNet-50, VGG-19, and DenseNet-121, on the AGGC22 dataset. Results show that Virchow2 and UNI achieve the highest F1-scores, while Prov-GigaPath performs best on rare Gleason grades. These findings highlight the potential of foundation models for improving WSI-based cancer classification.
Benchmarking Feature Extractors for Prostate Cancer Detection Using Graph Attention Networks: A Focus on Foundation Models
Akebli H.;Della Mea V.;Roitero K.
2025-01-01
Abstract
Accurate prostate cancer detection from Whole-Slide Images (WSIs) is critical for improving diagnostics. This study benchmarks 10 foundation models as feature extractors for graph-based classification using Graph Attention Networks (GATs). We compare them to CNN-based extractors, including ResNet-50, VGG-19, and DenseNet-121, on the AGGC22 dataset. Results show that Virchow2 and UNI achieve the highest F1-scores, while Prov-GigaPath performs best on rare Gleason grades. These findings highlight the potential of foundation models for improving WSI-based cancer classification.File in questo prodotto:
Non ci sono file associati a questo prodotto.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
